斯隆奖得主:当深度学习和博弈论相结合,能处理哪些社会问题?
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">5.10</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">知识分子</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">The Intellectual</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/0aa523ceb9684d78a15a1a0faa0ea965~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1728801435&x-signature=aW1dw3ehrNtxnMaA9pXAuJxIthM%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">算法与社会福祉的冲突是天然存在的吗?| 图源:pixabay.com</span></span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">导 读</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">今日,人工智能算法<span style="color: black;">已然</span>渗透到社会生活的方方面面:购物软件依靠算法判断消费者的消费习惯,推送特定商品;打车软件<span style="color: black;">经过</span>算法派单和定价,调配司机运力;社交软件利用算法分析关键词、推送<span style="color: black;">宣传</span>和内容。在这个过程中,尽可能为<span style="color: black;">商场</span>企业产生收益的算法,给社会生活带来便利的<span style="color: black;">同期</span>,<span style="color: black;">亦</span>因隐私、<span style="color: black;">卑视</span>等<span style="color: black;">许多</span>问题饱受诟病。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">然则</span>,算法与社会福祉的冲突是天然存在的吗?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">卡耐基梅隆大学(CMU)助理教授方飞认为,找对<span style="color: black;">科研</span>方向,算法<span style="color: black;">亦</span>能够<span style="color: black;">帮忙</span><span style="color: black;">处理</span><span style="color: black;">有些</span>社会性的问题。她的工作,<span style="color: black;">便是</span>将博弈论、人工智能与现实问题相结合,解决现实中的<span style="color: black;">繁杂</span>问题,<span style="color: black;">例如</span><span style="color: black;">守护</span>轮渡免受恐怖分子袭击,防止盗猎者在<span style="color: black;">守护</span>区内<span style="color: black;">损伤</span>动物,将即将过期的<span style="color: black;">食品</span>分给需要的人等等。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">方飞的工作扎实、稳重,在<span style="color: black;">行业</span>内赢得了不少赞誉。2020年,她入选IEEE“AI十大<span style="color: black;">潜能</span><span style="color: black;">名人</span>”;2021年<span style="color: black;">得到</span>IJCAI(国际人工智能联合会议)计算机与思想奖;2022年2月,她<span style="color: black;">得到</span>斯隆研究奖,一个奖励职业生涯<span style="color: black;">初期</span>杰出<span style="color: black;">青年</span>学者的奖项。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">方飞<span style="color: black;">此刻</span>CMU计算机科学学院软件<span style="color: black;">科研</span>所任职。她认为,<span style="color: black;">有些</span><span style="color: black;">状况</span>下,企业<span style="color: black;">经过</span>算法在最大化社会总收益时<span style="color: black;">亦</span>能得到较高的利润。而相比于人工智能的其他强应用性<span style="color: black;">行业</span>,AI向善还是一片“蓝海”。她<span style="color: black;">期盼</span>有<span style="color: black;">更加多</span><span style="color: black;">科研</span>者能加入这个<span style="color: black;">行业</span>,为<span style="color: black;">提高</span>整体社会福利作出贡献。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">她说,进入这个<span style="color: black;">行业</span>最重要的,是要有<span style="color: black;">发掘</span>问题的眼睛。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">以下是《知识分子》和方飞的对话,行文有删减。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">撰文 | 王一苇</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">责编 | 陈晓雪</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">● </span><span style="color: black;">● </span><span style="color: black;">●</span></span></p>
<div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-qvj2lq49k0/4f399d429ab947f9a0c7cbe3c8ed2284~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1728801435&x-signature=HvtJPty4ZPWbWcZNgUGTfAaT7Iw%3D" style="width: 50%; margin-bottom: 20px;"></div>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">方飞,卡耐基梅隆大学(CMU)助理教授,斯隆<span style="color: black;">科研</span>奖<span style="color: black;">得到</span>者</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">当人工智能与博弈论相结合</span></span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:博弈论是1944年的时候冯·诺依曼和奥斯卡·摩根斯特恩提出的,跟人工智能结合的博弈论<span style="color: black;">科研</span>是什么时候、怎么发展起来的?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:一<span style="color: black;">起始</span>,博弈论<span style="color: black;">重点</span>在经济<span style="color: black;">行业</span>里面</span><span style="color: black;">(发展)</span><span style="color: black;">,诺贝尔经济学奖得主经常是<span style="color: black;">科研</span>博弈论的。<span style="color: black;">因为</span>计算机的发展,<span style="color: black;">大众</span>很感兴趣在博弈论里什么样的问题是可计算的,或是能有效地计算出结果的,就会去<span style="color: black;">科研</span><span style="color: black;">区别</span>的激励形式和<span style="color: black;">区别</span>的<span style="color: black;">平衡</span>,看什么样的<span style="color: black;">平衡</span>概念在什么样的博弈中是有效的算法,什么样的问题是NP hard——比较难解的。有了预测,就<span style="color: black;">起始</span>处理计算上的挑战。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">安全博弈<span style="color: black;">关联</span>的<span style="color: black;">发展</span>是在2006年,Tuomas Sandholm和Vincent Conitzer两个大牛在Economics and Computation</span><span style="color: black;">(经济学与计算)</span><span style="color: black;">会议上发了一篇论文</span><span style="color: black;"> </span><span style="color: black;">,说在斯坦伯格<span style="color: black;">平衡</span><span style="color: black;">其中</span>,在比较简单的问题上,是<span style="color: black;">能够</span>有多项式时间的算法可解的,<span style="color: black;">然则</span>当这个问题延伸到有一方博弈参与者的支付函数或在博弈中的收益有多种可能、类型未知时,这个问题就不可解。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">在这之后,我的导师Milind Tambe是最先把博弈论应用到安全<span style="color: black;">行业</span>的。他<span style="color: black;">科研</span>的是在一个<span style="color: black;">详细</span>的问题里可不<span style="color: black;">能够</span>应用斯坦伯格博弈</span><span style="color: black;">(Stackelberg Game,指存在两方的完全信息动态博弈,双方都是<span style="color: black;">按照</span>对方可能的策略来<span style="color: black;">选取</span>自己的策略)</span><span style="color: black;">去分析。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">例如</span>,<span style="color: black;">咱们</span>想<span style="color: black;">守护</span>洛杉矶的机场。在<span style="color: black;">守护</span>机场的问题里,有<span style="color: black;">守护</span>者<span style="color: black;">亦</span>有攻击者。机场的巡护人员,<span style="color: black;">亦</span><span style="color: black;">便是</span>防御的这一方,<span style="color: black;">每日</span>都要进行巡逻。<span style="color: black;">那样</span><span style="color: black;">做为</span>攻击者,可能会花很长的时间去观察巡逻人员的巡逻有什么样的规律,是不是<span style="color: black;">每一个</span>周一都在一号航站楼巡逻,<span style="color: black;">每一个</span>周二都在二号航站楼巡逻之类的,<span style="color: black;">而后</span><span style="color: black;">按照</span>观察到的这些规律去找巡逻<span style="color: black;">方法</span>里相对的弱点,避免被巡逻人员抓到。<span style="color: black;">她们</span>会有<span style="color: black;">各样</span>各样的攻击,<span style="color: black;">例如</span>说有人会把手枪、步枪等<span style="color: black;">各样</span>各样的武器带进机场。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">巡逻人员一方面会在进入机场的多个路口设<span style="color: black;">检测</span>站,另一方面在各个航站楼里面用警犬巡逻,<span style="color: black;">然则</span><span style="color: black;">由于</span><span style="color: black;">检测</span>站的数量以及警犬的数量都<span style="color: black;">不足</span>多,<span style="color: black;">因此</span>需要去决定<span style="color: black;">每日</span>去什么样的<span style="color: black;">地区</span>做什么。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">由于</span>有<span style="color: black;">这般</span>的不对<span style="color: black;">叫作</span>关系,<span style="color: black;">她们</span>把这个问题建模成一个斯坦伯格博弈:巡逻者这一方是领导者,先去<span style="color: black;">选取</span>一个巡逻策略;而攻击者那方是跟随者,观察领导者的策略以后,再去<span style="color: black;">回复</span>,<span style="color: black;">选取</span>更好的攻击方式<span style="color: black;">回复</span>现有的巡逻<span style="color: black;">方法</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">在这个<span style="color: black;">基本</span>上,<span style="color: black;">她们</span>又<span style="color: black;">研发</span>了一系列的模型和算法来去<span style="color: black;">科研</span>这个<span style="color: black;">行业</span>,慢慢<span style="color: black;">作为</span>了最<span style="color: black;">起始</span>的一波安全博弈项目。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">我加入Tambe的<span style="color: black;">科研</span>组以后,<span style="color: black;">起始</span>应用<span style="color: black;">更加多</span>的<span style="color: black;">设备</span>学习手段。<span style="color: black;">咱们</span><span style="color: black;">起始</span><span style="color: black;">科研</span>移动的<span style="color: black;">目的</span>,<span style="color: black;">例如</span>轮渡是不是<span style="color: black;">亦</span><span style="color: black;">能够</span>用类似的模型建模,以及能<span style="color: black;">不可</span>找到快速的算法来去求解最好的巡逻方式。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">美国的海防在各个渡口都有巡逻。在纽约,巡逻人员其中一部分任务<span style="color: black;">便是</span>要<span style="color: black;">守护</span>曼哈顿岛到史丹顿岛的轮渡。这个问题里,轮渡<span style="color: black;">便是</span>移动的<span style="color: black;">目的</span>,而巡逻人员<span style="color: black;">亦</span>驾着小船在<span style="color: black;">一直</span>移动,<span style="color: black;">咱们</span>想要做的事情<span style="color: black;">便是</span>用博弈论建模,并且去构建一个算法,找到最佳的路线<span style="color: black;">选取</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">最<span style="color: black;">起始</span>的模型<span style="color: black;">重点</span><span style="color: black;">便是</span>两方的斯坦伯格博弈,后来<span style="color: black;">亦</span>有<span style="color: black;">科研</span>多方的。前两年的<span style="color: black;">有些</span><span style="color: black;">守护</span>动物<span style="color: black;">关联</span>的工作里,<span style="color: black;">守护</span>区的人跟<span style="color: black;">咱们</span>说,有时候会有居民告诉<span style="color: black;">她们</span>,听说盗猎者去哪里盗猎了<span style="color: black;">或</span>是将要去哪里盗猎。当时现有的模型都<span style="color: black;">无</span>办法去<span style="color: black;">思虑</span><span style="color: black;">这般</span>的<span style="color: black;">原因</span>,<span style="color: black;">咱们</span>就<span style="color: black;">创立</span>新的模型来<span style="color: black;">思虑</span>多于两个博弈者的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:听上去这个<span style="color: black;">行业</span>实用性非常强。<span style="color: black;">那样</span>博弈论跟AI结合,想要完成的科学<span style="color: black;">目的</span>是什么?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:在我看来,它的科学<span style="color: black;">目的</span><span style="color: black;">便是</span><span style="color: black;">期盼</span>能够去找到更有效的、更快速的算法来<span style="color: black;">处理</span>更<span style="color: black;">繁杂</span>或更<span style="color: black;">实质</span>的博弈。博弈论是理论框架,而AI或深度学习是工具,在这个框架下去求解数学问题。<span style="color: black;">全部</span>计算博弈论<span style="color: black;">行业</span>,<span style="color: black;">大众</span>都在做的事情<span style="color: black;">便是</span><span style="color: black;">期盼</span>有更好更快的算法,能够去求解更<span style="color: black;">繁杂</span>的博弈。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">非常多</span>时候,计算和博弈论结合<span style="color: black;">没法</span>找到快速的解。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">一种<span style="color: black;">状况</span>是这个博弈太<span style="color: black;">繁杂</span>了,<span style="color: black;">无</span>办法用数学规划求解,<span style="color: black;">由于</span><span style="color: black;">倘若</span>用数据规划去求解的话,需要几百万的电量、几百万的限制<span style="color: black;">要求</span>,<span style="color: black;">倘若</span>真的在<span style="color: black;">设备</span>上算,光去求解,这个<span style="color: black;">设备</span>的内存就爆掉了。<span style="color: black;">这般</span>的<span style="color: black;">状况</span>下,<span style="color: black;">咱们</span><span style="color: black;">能够</span><span style="color: black;">经过</span>深度学习的方式,找到这个大问题里面一个较优的解,<span style="color: black;">或</span>说比较靠近<span style="color: black;">平衡</span>的策略。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">第二种<span style="color: black;">状况</span>,有时候在博弈问题里需要<span style="color: black;">思虑</span>人的<span style="color: black;">行径</span>。最传统的博弈论假设所有的人都是理性人,但后来提出,并不是所有人都是完全理性的,<span style="color: black;">此时</span>候就要去理解人的<span style="color: black;">行径</span>模式。<span style="color: black;">设备</span>学习<span style="color: black;">能够</span><span style="color: black;">帮忙</span><span style="color: black;">咱们</span>,<span style="color: black;">经过</span>人以往的<span style="color: black;">行径</span>数据来给人的<span style="color: black;">行径</span>建模,在<span style="color: black;">这般</span>的模型下去再去找博弈中其他博弈方的解。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">第三种<span style="color: black;">状况</span>是逆向博弈论</span><span style="color: black;">(inversive game theory)</span><span style="color: black;">。<span style="color: black;">通常</span>的博弈论的问题是说,我告诉你博弈是这几个人在进行,在<span style="color: black;">区别</span>的<span style="color: black;">状况</span>下,<span style="color: black;">每一个</span>人的收益会是多少,<span style="color: black;">而后</span>问你<span style="color: black;">针对</span><span style="color: black;">这般</span>一个博弈问题,<span style="color: black;">平衡</span>策略应该是什么。逆向博弈论<span style="color: black;">便是</span>反过来,我能够观察到<span style="color: black;">大众</span>在博弈中采取什么样的行动,问能<span style="color: black;">不可</span>去找到<span style="color: black;">每一个</span>人的收益函数是什么。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">在<span style="color: black;">这般</span>一个问题里,有<span style="color: black;">非常多</span>观察数据来描述<span style="color: black;">每一个</span>人的<span style="color: black;">行径</span>和他采取的行动,<span style="color: black;">咱们</span>要去从行动里逆推收益函数,这个时候<span style="color: black;">设备</span>学习<span style="color: black;">亦</span><span style="color: black;">能够</span><span style="color: black;">帮忙</span><span style="color: black;">咱们</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">我提的这三个是我自己做过的,都是把<span style="color: black;">设备</span>学习<span style="color: black;">做为</span>工具来<span style="color: black;">处理</span>博弈论里面的问题。还有其他结合博弈论和深度学习的方式。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">反过来,<span style="color: black;">亦</span><span style="color: black;">能够</span>用博弈论去<span style="color: black;">处理</span><span style="color: black;">设备</span>学习里面的<span style="color: black;">非常多</span>问题。一个例子是卡耐基梅隆大学招学生,最初一轮筛选中,<span style="color: black;">每一个</span>人把申请资料交过来,<span style="color: black;">大众</span>来看。事实上除了人做判断以外,学校有一个算法来判断这个学生是不是值得<span style="color: black;">思虑</span>的,防止有<span style="color: black;">有些</span>被人漏掉。这两条线是完全独立地在做判断,<span style="color: black;">便是</span>说只要人<span style="color: black;">或</span><span style="color: black;">设备</span>认为这个学生应该值得被<span style="color: black;">思虑</span>,<span style="color: black;">那样</span>就会进入下一轮。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">能够</span>想象,在<span style="color: black;">这般</span>的<span style="color: black;">状况</span>下,<span style="color: black;">倘若</span>学生<span style="color: black;">晓得</span>你是用什么样的方式来进行筛选,<span style="color: black;">她们</span>可能会去想办法改变自己的<span style="color: black;">有些</span><span style="color: black;">行径</span><span style="color: black;">或</span>分数,<span style="color: black;">按照</span>算法去<span style="color: black;">调节</span>自己的<span style="color: black;">行径</span>。在<span style="color: black;">这般</span>的<span style="color: black;">状况</span>下,<span style="color: black;">咱们</span>其实<span style="color: black;">能够</span>用博弈论去分析</span><span style="color: black;">(如何<span style="color: black;">处理</span>这个问题)</span><span style="color: black;">。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:博弈论和<span style="color: black;">设备</span>学习的结合<span style="color: black;">能够</span><span style="color: black;">处理</span><span style="color: black;">那些</span>类型的社会问题?你<span style="color: black;">通常</span>是怎么去找到这些现实社会中的问题和痛点的?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:<span style="color: black;">咱们</span>帮美国海防设计巡逻路线,帮动物<span style="color: black;">守护</span>组织设计护林员的巡护路线。前者是安全性问题,后者与环境的可<span style="color: black;">连续</span>发展<span style="color: black;">关联</span>。还有移动性</span><span style="color: black;">(mobility)</span><span style="color: black;"><span style="color: black;">关联</span>的问题,<span style="color: black;">例如</span>交通运输等,<span style="color: black;">咱们</span><span style="color: black;">亦</span>在<span style="color: black;">科研</span>中。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">有一部分是<span style="color: black;">各样</span>各样<span style="color: black;">原因</span>的巧合叠加,另一部分其实是<span style="color: black;">由于</span><span style="color: black;">咱们</span>之前做过的<span style="color: black;">有些</span>东西,延展到了类似的或是<span style="color: black;">关联</span>的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">例如</span>说我博一的时候做了<span style="color: black;">守护</span>轮渡的工作。<span style="color: black;">咱们</span>讲演、发了paper之后,去跟<span style="color: black;">区别</span>的人<span style="color: black;">交流</span>,有人就<span style="color: black;">说到</span>,你们这个问题听起来跟动物<span style="color: black;">守护</span>护林员的巡逻问题挺类似的,有<span style="color: black;">无</span><span style="color: black;">思虑</span>过把你们的算法应用到那个问题上。这给了<span style="color: black;">咱们</span>动力,去跟<span style="color: black;">关联</span>的专家去<span style="color: black;">交流</span>,<span style="color: black;">认识</span><span style="color: black;">实质</span>的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">我来CMU之后,<span style="color: black;">咱们</span>还做一个新的系列工作,是<span style="color: black;">食品</span>救助。我一<span style="color: black;">起始</span>不<span style="color: black;">晓得</span>匹兹堡有<span style="color: black;">这般</span>一个非营利组织,是有一次在<span style="color: black;">咱们</span>学校会议室里面正好遇到了这个组织的CEO,互发了名片。我去查了一下<span style="color: black;">她们</span>的组织在干什么,<span style="color: black;">发掘</span><span style="color: black;">尤其</span>有意思。后来我自己<span style="color: black;">亦</span>去给他们做志愿者,又去跟<span style="color: black;">她们</span>交流,跟<span style="color: black;">她们</span>说<span style="color: black;">咱们</span>做了<span style="color: black;">非常多</span>AI<span style="color: black;">关联</span>的工作,<span style="color: black;">期盼</span><span style="color: black;">瞧瞧</span><span style="color: black;">咱们</span>的技术有什么能够应用到这个问题里面,能够<span style="color: black;">帮忙</span>你们平台发展得更好。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">从<span style="color: black;">这般</span>的讨论<span style="color: black;">起始</span>,慢慢的就有了越来越多真正的工作,<span style="color: black;">咱们</span><span style="color: black;">此刻</span>在这个方向上<span style="color: black;">已然</span>发了三四篇paper,(算法)<span style="color: black;">亦</span>在<span style="color: black;">她们</span>的系统里面<span style="color: black;">已然</span>用上了,这<span style="color: black;">亦</span>是我<span style="color: black;">尤其</span>高兴的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">食品</span>救助分两个部分,<span style="color: black;">第1</span>个部分是匹配donor和receiver,<span style="color: black;">便是</span><span style="color: black;">食品</span>捐助者和<span style="color: black;">食品</span>接受者;第二个部分是匹配志愿者去运送<span style="color: black;">食品</span>。<span style="color: black;">例如</span>我当时去做志愿者,CMU的一个咖啡厅有剩下的蛋糕,<span style="color: black;">她们</span>想捐给<span style="color: black;">周边</span>的单身母亲救助站,就在手机APP上说<span style="color: black;">咱们</span>需要志愿者来把<span style="color: black;">食品</span>从CMU的咖啡厅送到救助站去,我一看很高兴,就在我楼上,就赶紧接了这个单,把<span style="color: black;">食品</span>送过去了。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">咱们</span><span style="color: black;">重点</span><span style="color: black;">科研</span>的是平台和志愿者之间怎么匹配的问题,<span style="color: black;">或</span>说怎么能够<span style="color: black;">帮忙</span>平台更好的找到能够来运送<span style="color: black;">食品</span>的志愿者,<span style="color: black;">怎么样</span>让志愿者在这个过程中<span style="color: black;">得到</span>更好的体验。在这个问题里,<span style="color: black;">咱们</span>的<span style="color: black;">目的</span><span style="color: black;">便是</span>能够不要发太多的的<span style="color: black;">通告</span>,<span style="color: black;">由于</span><span style="color: black;">倘若</span>每来一个单就给所有的志愿者发<span style="color: black;">通告</span>的话</span><span style="color: black;">(会很打扰)</span><span style="color: black;">。<span style="color: black;">咱们</span><span style="color: black;">期盼</span>不要发太多的<span style="color: black;">通告</span>,<span style="color: black;">然则</span>又能够<span style="color: black;">提高</span>接单率和缩<span style="color: black;">短期</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">新的<span style="color: black;">举荐</span>机制上线以后,总的接单率<span style="color: black;">提升</span>了,平均接单时间减少了。这个问题其实跟打车软件有一点像,<span style="color: black;">亦</span>是有一个派单的过程,<span style="color: black;">然则</span>很不<span style="color: black;">同样</span>的是,这是一个纯志愿者的非营利性平台,给打车软件做的定价策略是完全<span style="color: black;">不可</span>用的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">找到真问题</span></span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:寻找<span style="color: black;">处理</span><span style="color: black;">方法</span>的过程中,你觉得最困难的事情是什么?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:是找问题。找能够利用人工智能、真的产生价值的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">像<span style="color: black;">食品</span>救助这个问题,我一<span style="color: black;">起始</span>做完志愿者之后,觉得<span style="color: black;">全部</span>系统里面有太多<span style="color: black;">能够</span>改进的点了。我当时写了一个很长的邮件给<span style="color: black;">她们</span>CEO,说<span style="color: black;">按照</span>我的经验,一二三四这几个方面都<span style="color: black;">能够</span>改进,<span style="color: black;">而后</span><span style="color: black;">咱们</span>的人工智能<span style="color: black;">能够</span>在<span style="color: black;">每一个</span>方面做点什么。我提的<span style="color: black;">意见</span>是,你们给志愿者的<span style="color: black;">指点</span>信息太模糊了,需要<span style="color: black;">经过</span><span style="color: black;">有些</span>智能的方式<span style="color: black;">供给</span>更清楚的<span style="color: black;">指点</span>,<span style="color: black;">例如</span>应该去找谁跟谁对接,在什么<span style="color: black;">地区</span>跟别人见面之类的,<span style="color: black;">另一</span><span style="color: black;">能够</span>让一个志愿者送多个单。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">后来我去跑到<span style="color: black;">她们</span>办公室里面去跟<span style="color: black;">她们</span>聊,<span style="color: black;">她们</span>就说你提的问题都<span style="color: black;">特别有</span>意思,但不是<span style="color: black;">咱们</span>真正关心的点。<span style="color: black;">针对</span><span style="color: black;">她们</span>的组织<span style="color: black;">来讲</span>,需要看的是更大的方面,<span style="color: black;">她们</span>真正的痛点是,怎么能够让<span style="color: black;">更加多</span>的志愿者来参与,让<span style="color: black;">食品</span>不被浪费,能够<span style="color: black;">即时</span>地送到该送的人手里面。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">因此</span>要<span style="color: black;">持续</span>地跟<span style="color: black;">她们</span>讨论,我的学生<span style="color: black;">亦</span>去<span style="color: black;">她们</span>那边实习了两天,帮<span style="color: black;">她们</span>去做派单的工作,体验一下从内部的<span style="color: black;">方向</span>来看派单的过程是什么样子。在<span style="color: black;">持续</span>的讨论中,最后才慢慢找到了一个对<span style="color: black;">她们</span><span style="color: black;">来讲</span>是非常关心的,<span style="color: black;">针对</span><span style="color: black;">咱们</span><span style="color: black;">来讲</span>是人工智能真的<span style="color: black;">能够</span>发挥作用的问题。在我看来这可能是最花时间的一个<span style="color: black;">过程</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">另一个我觉得<span style="color: black;">尤其</span>有挑战的<span style="color: black;">便是</span>,怎么能够让<span style="color: black;">她们</span>真的去用你做的这个东西。可能他把数据给你了,<span style="color: black;">而后</span>你做的这个东西看起来挺有趣的,发了论文了,<span style="color: black;">然则</span>他觉得,我<span style="color: black;">无</span>能力去把你这个算法真的实<span style="color: black;">此刻</span>系统里面,<span style="color: black;">咱们</span>太忙了,有其他的事情要做,这个东西可能不是<span style="color: black;">她们</span>的最高优先级。<span style="color: black;">倘若</span>说是一个纯<span style="color: black;">商场</span>的问题,这个工具<span style="color: black;">能够</span>给你的利润<span style="color: black;">加强</span>10个点,<span style="color: black;">她们</span>可能就很积极地去做了,<span style="color: black;">然则</span>公益组织不是<span style="color: black;">这般</span>的,<span style="color: black;">她们</span>有其他方面的<span style="color: black;">思虑</span>。怎么能够说服<span style="color: black;">她们</span>去做<span style="color: black;">实质</span>的测试和应用,在<span style="color: black;">咱们</span>看来<span style="color: black;">亦</span>非是常难的一个点。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">咱们</span>的经验<span style="color: black;">便是</span>,<span style="color: black;">首要</span>要充分尊重<span style="color: black;">她们</span>的意愿,积极地跟<span style="color: black;">她们</span>沟通;其次是<span style="color: black;">尽可能</span>减少<span style="color: black;">她们</span>需要付出的成本,<span style="color: black;">包含</span>时间成本、人力成本以及可能的金钱成本。<span style="color: black;">咱们</span>在进行<span style="color: black;">实质</span>测试的时候,最好是把能干的活都干了,<span style="color: black;">她们</span>可能只需要花几个小时来跟<span style="color: black;">咱们</span>开个会,讨论一下<span style="color: black;">方法</span>,<span style="color: black;">而后</span><span style="color: black;">准许</span>一下<span style="color: black;">咱们</span>上线前的测试是过关了的,就<span style="color: black;">能够</span>上线。<span style="color: black;">这般</span>的话<span style="color: black;">她们</span>就更愿意去做<span style="color: black;">这般</span>的测试。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:你们的项目<span style="color: black;">好似</span><span style="color: black;">此刻</span>还是偏公益的<span style="color: black;">占多数</span><span style="color: black;">有些</span>,是<span style="color: black;">思虑</span>到<span style="color: black;">科研</span>本身的性质,还是你觉得这是最大化效益的工作?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:<span style="color: black;">咱们</span>并<span style="color: black;">无</span>排斥<span style="color: black;">商场</span>化的项目,<span style="color: black;">然则</span>我个人非常<span style="color: black;">期盼</span>能够<span style="color: black;">帮忙</span><span style="color: black;">处理</span><span style="color: black;">有些</span>社会性的问题。我开玩笑说<span style="color: black;">咱们</span>在做高级志愿者,<span style="color: black;">帮忙</span><span style="color: black;">有些</span>政府组织或工业公益组织更好地向社会<span style="color: black;">供给</span>更高质量的服务。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">这些问题是对社会<span style="color: black;">来讲</span>是非常重要的,它可能并不<span style="color: black;">必定</span>能够直接产生<span style="color: black;">商场</span>利益,<span style="color: black;">然则</span>这些都是影响<span style="color: black;">非常多</span>人的问题,<span style="color: black;">然则</span>又<span style="color: black;">无</span><span style="color: black;">非常多</span>人去做。<span style="color: black;">亦</span><span style="color: black;">能够</span>理解,<span style="color: black;">此刻</span>人工智能这么火,<span style="color: black;">能够</span>想象去业界的话能赚<span style="color: black;">非常多</span>钱,<span style="color: black;">那样</span>当你有这个能力的时候,可能<span style="color: black;">非常多</span>人<span style="color: black;">选取</span>了去<span style="color: black;">挣钱</span>。<span style="color: black;">亦</span>许<span style="color: black;">非常多</span>人是对这件事感兴趣的,<span style="color: black;">然则</span>真正投入去做的人<span style="color: black;">无</span><span style="color: black;">那样</span>多。我<span style="color: black;">亦</span><span style="color: black;">期盼</span><span style="color: black;">咱们</span>能够一方面自己去做<span style="color: black;">有些</span>工作,另一方面<span style="color: black;">亦</span>能够让<span style="color: black;">更加多</span>的人去参与<span style="color: black;">这般</span>的工作。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">我在学校里教AI for social good</span><span style="color: black;">(人工智能向善)</span><span style="color: black;">的课,<span style="color: black;">亦</span>是<span style="color: black;">期盼</span>能培养<span style="color: black;">更加多</span>的学生去接触<span style="color: black;">这般</span>的问题,<span style="color: black;">处理</span><span style="color: black;">这般</span>的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:你<span style="color: black;">刚才</span><span style="color: black;">亦</span>说了,找问题是一个难点。你怎么看这个<span style="color: black;">行业</span><span style="color: black;">将来</span>的发展,<span style="color: black;">将来</span>它还能够应用在<span style="color: black;">更加多</span>的<span style="color: black;">行业</span>、<span style="color: black;">更加多</span>的问题上吗?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:我觉得有<span style="color: black;">非常多</span>值得做的问题。我过去三年都在做AAAI</span><span style="color: black;">(人工智能促进协会)</span><span style="color: black;">的AI for social impact special track</span><span style="color: black;">(AI社会影响力专刊)</span><span style="color: black;">的co-chair</span><span style="color: black;">(联合主席)</span><span style="color: black;">,<span style="color: black;">咱们</span>每年<span style="color: black;">亦</span>都收到不少的论文,能看到<span style="color: black;">非常多</span>人对这个感兴趣,<span style="color: black;">亦</span><span style="color: black;">持续</span>地在做这方面的工作。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">将来</span>还是有<span style="color: black;">非常多</span><span style="color: black;">能够</span>做的问题的,<span style="color: black;">例如</span>像联合国的17个可<span style="color: black;">连续</span>发展的<span style="color: black;">目的</span>,<span style="color: black;">咱们</span><span style="color: black;">此刻</span>的工作可能只<span style="color: black;">触及</span>到其中的四个<span style="color: black;">目的</span>,其实还有<span style="color: black;">非常多</span>其他<span style="color: black;">目的</span>,中间可能<span style="color: black;">亦</span><span style="color: black;">触及</span>到许多值得去做的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">社会总收益 vs 企业利润</span></span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:<span style="color: black;">非常多</span>企业用AI算法<span style="color: black;">提高</span>效率,但<span style="color: black;">亦</span>受到<span style="color: black;">有些</span>诟病。外卖企业里,骑手<span style="color: black;">由于</span>这些算法的规定,<span style="color: black;">或</span>企业本身的策略,福利是在<span style="color: black;">持续</span>下降的。你对<span style="color: black;">关联</span>的AI应用有<span style="color: black;">那些</span>观察,在企业责任<span style="color: black;">或</span>用AI算法来改善的方面,有<span style="color: black;">那些</span>能做的事情吗?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:<span style="color: black;">机构</span>肯定要追求利润的,这肯定是<span style="color: black;">她们</span>的<span style="color: black;">重点</span><span style="color: black;">目的</span>。<span style="color: black;">那样</span>除了利润之外,博弈论<span style="color: black;">或</span>机制设计里面经常会谈到一个 social welfare,社会福利,所有人的收益之和。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">在外卖平台或打车平台,平台的收益加上骑手<span style="color: black;">或</span>是司机以及等餐的人、想要打车的人,所有人的收益之和是需要关注的一个<span style="color: black;">目的</span>函数。<span style="color: black;">咱们</span>之前做优步等打车平台的定价策略的<span style="color: black;">科研</span>的时候,<span style="color: black;">便是</span>把<span style="color: black;">目的</span>设为了最大化社会收益。怎么计算社会收益?<span style="color: black;">例如</span><span style="color: black;">针对</span>乘客<span style="color: black;">来讲</span>,他很想去这个<span style="color: black;">地区</span>,他愿意花100块钱去,<span style="color: black;">而后</span>他付了50块钱,<span style="color: black;">那样</span>他的收益<span style="color: black;">便是</span>100-50。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">咱们</span>很难去跟<span style="color: black;">商场</span><span style="color: black;">机构</span>说,你们不要关注利润了,来关注社会总收益,这是非常困难的。<span style="color: black;">然则</span><span style="color: black;">咱们</span>看到的是,什么样的<span style="color: black;">状况</span>下,<span style="color: black;">大众</span>会更关注除了利润以外的<span style="color: black;">目的</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">一个是监管,<span style="color: black;">例如</span>保险<span style="color: black;">机构</span>,可能一<span style="color: black;">起始</span>的时候,当<span style="color: black;">无</span>足够的监管的时候,保险<span style="color: black;">机构</span>的定价策略会是有非常有偏见的,当<span style="color: black;">有些</span>事情被爆出来之后监管介入,监管可能就<span style="color: black;">需求</span>你保险的定价策略更公平。<span style="color: black;">那样</span>在<span style="color: black;">这般</span>的<span style="color: black;">状况</span>下,这些<span style="color: black;">机构</span>就会更<span style="color: black;">思虑</span>除了利润以外的社会总收益,<span style="color: black;">例如</span>公平性的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">另一个是,<span style="color: black;">有些</span><span style="color: black;">状况</span>下,<span style="color: black;">咱们</span><span style="color: black;">能够</span>去向这些<span style="color: black;">机构</span>证明,在你们的问题里,其实最大化社会收总收益是跟最大化总利润是差别非常小的。<span style="color: black;">倘若</span>你去最大化社会总收益,<span style="color: black;">亦</span>能够<span style="color: black;">得到</span>比<span style="color: black;">很强</span>的收益,可能不<span style="color: black;">必定</span>是最大化,<span style="color: black;">然则</span><span style="color: black;">亦</span>比较接近,但它能够在社会总福利方面有一个更大的<span style="color: black;">提高</span>。<span style="color: black;">咱们</span><span style="color: black;">科研</span>打车平台的时候做过分析,在<span style="color: black;">必定</span>的假设之下,最大化社会总收益时<span style="color: black;">亦</span>能够得到比较高的利润。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;"><span style="color: black;">怎样</span>消除数据偏见?</span></span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:在AI<span style="color: black;">科研</span>里,数据是<span style="color: black;">基本</span>,<span style="color: black;">然则</span><span style="color: black;">亦</span>经常会遇到<span style="color: black;">有些</span>数据不准确<span style="color: black;">或</span>偏见的<span style="color: black;">状况</span>,你的<span style="color: black;">科研</span>会遇到类似的问题吗?怎么去<span style="color: black;">处理</span>这些问题?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:会。拿动物<span style="color: black;">守护</span>的<span style="color: black;">状况</span><span style="color: black;">来讲</span>,<span style="color: black;">咱们</span>收集的数据就有<span style="color: black;">非常多</span>的问题。<span style="color: black;">例如</span>说现有的数据都是护林员们之前非常辛苦地去巡逻,收集来的。<span style="color: black;">然则</span><span style="color: black;">她们</span>并不是覆盖了所有的区域,可能某些区域去的比较多,另<span style="color: black;">有些</span>区域去的比较少,这就<span style="color: black;">引起</span>去的比较多的区域,可能有<span style="color: black;">更加多</span>的数据,<span style="color: black;">况且</span>这个<span style="color: black;">地区</span>到底盗猎有多高发,是比较准确的,能有一个相对比较准确的估计;<span style="color: black;">然则</span>对<span style="color: black;">哪些</span>去的比较少的<span style="color: black;">地区</span>,即使<span style="color: black;">她们</span>说我去了,我<span style="color: black;">无</span>找到猎套,并不<span style="color: black;">寓意</span>着那个地方从来<span style="color: black;">无</span><span style="color: black;">出现</span>过盗猎,可能只是<span style="color: black;">由于</span><span style="color: black;">她们</span>去的少,去的时候正好没<span style="color: black;">发掘</span>猎套。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">况且</span><span style="color: black;">倘若</span>看总体的数据量,<span style="color: black;">她们</span>去过的所有<span style="color: black;">地区</span>里面找到猎套的点肯定是少的,还是有<span style="color: black;">更加多</span>的时候<span style="color: black;">她们</span>在路上走,什么<span style="color: black;">亦</span><span style="color: black;">无</span><span style="color: black;">发掘</span>。这<span style="color: black;">亦</span>说明数据中的不平衡,<span style="color: black;">亦</span>是<span style="color: black;">咱们</span>在设计<span style="color: black;">设备</span>学习算法来学习盗猎者的<span style="color: black;">行径</span>模型的时候需要去处理的挑战。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">咱们</span>尝试了<span style="color: black;">各样</span>各样的办法,像最初的一版算法里面,<span style="color: black;">咱们</span>把<span style="color: black;">全部</span><span style="color: black;">守护</span>区分成多个地块,<span style="color: black;">倘若</span>这个地块它的数据量比<span style="color: black;">很强</span>,<span style="color: black;">能够</span>用稍微<span style="color: black;">繁杂</span>一点的<span style="color: black;">设备</span>学习算法;<span style="color: black;">倘若</span>这个地块数据量<span style="color: black;">不足</span>,<span style="color: black;">咱们</span>就把数据量<span style="color: black;">不足</span>的<span style="color: black;">哪些</span>地块所有的数据<span style="color: black;">所有</span>结合起来,<span style="color: black;">而后</span>去找一个<span style="color: black;">针对</span>数据量<span style="color: black;">需求</span><span style="color: black;">无</span><span style="color: black;">那样</span>高的<span style="color: black;">设备</span>学习算法,<span style="color: black;">例如</span>决策树之类的<span style="color: black;">办法</span>去预测。这是一<span style="color: black;">起始</span>的想法。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">后来<span style="color: black;">咱们</span>又做了<span style="color: black;">非常多</span>其他的尝试,<span style="color: black;">例如</span>在黄泥河<span style="color: black;">守护</span>区</span><span style="color: black;">(<span style="color: black;">位置于</span>吉林省)</span><span style="color: black;">,<span style="color: black;">咱们</span>给当地的护林员发过问卷,问<span style="color: black;">她们</span>在<span style="color: black;">咱们</span>划分出来的<span style="color: black;">全部</span><span style="color: black;">守护</span>区的多个区域里,<span style="color: black;">每一个</span>区域的总体盗猎<span style="color: black;">危害</span>是高还是低,<span style="color: black;">按照</span><span style="color: black;">她们</span>的回答,<span style="color: black;">咱们</span>去额外采样<span style="color: black;">有些</span>新的数据点放到<span style="color: black;">咱们</span>的数据集里面,这一方面能够<span style="color: black;">帮忙</span><span style="color: black;">咱们</span><span style="color: black;">增多</span><span style="color: black;">更加多</span>的数据,另一方面<span style="color: black;">亦</span><span style="color: black;">能够</span>纠正这些护林员本身的偏见。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">人工智能进入死胡同了吗?</span></span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:<span style="color: black;">近期</span>有一种说法是,AI<span style="color: black;">已然</span>走进了一个死胡同,只适合处理<span style="color: black;">有些</span>存在低<span style="color: black;">危害</span>、存在完美答案的问题。你怎么看这个说法?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:一方面,咱们要承认<span style="color: black;">此刻</span>人工智能还是<span style="color: black;">无</span><span style="color: black;">那样</span>高级的,是真的<span style="color: black;">无</span><span style="color: black;">那样</span>厉害。<span style="color: black;">然则</span>我觉得还是有<span style="color: black;">非常多</span><span style="color: black;">能够</span>做的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">一个是高<span style="color: black;">危害</span>的问题。相对来讲,在高<span style="color: black;">危害</span>问题上,人工智能是辅助人类进行决策的,并不是想要代替人类进行决策,<span style="color: black;">咱们</span><span style="color: black;">晓得</span>决策是高<span style="color: black;">危害</span>的决策,<span style="color: black;">咱们</span>做的事情是<span style="color: black;">期盼</span>能够给决策者<span style="color: black;">供给</span><span style="color: black;">更加多</span>信息及<span style="color: black;">更加多</span>的可选<span style="color: black;">方法</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">此刻</span>有<span style="color: black;">非常多</span>人在做可解释性的人工智能,<span style="color: black;">必定</span>程度上是想要<span style="color: black;">处理</span>在高<span style="color: black;">危害</span>场景应用AI的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">其中一个方向是,最后我不要一个神经网络,我需要的是一个决策树,是能够画出来、人能够直接看得懂的,一个基于规则的<span style="color: black;">归类</span>器。可能它的表现<span style="color: black;">无</span>深度学习<span style="color: black;">那样</span>好,但还是<span style="color: black;">能够</span>有比较好的表现。在我去训练决策树的过程中,可能还是需要深度学习的,只是最后呈现的是一个决策树,<span style="color: black;">这般</span>的话人<span style="color: black;">最少</span>能够看得懂。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">还有<span style="color: black;">有些</span>其他路线,<span style="color: black;">例如</span>说我预测或做决策还是用深度学习的模型,<span style="color: black;">然则</span>我去跟人用自然语言解释它在做出<span style="color: black;">这般</span>的预测和决策时,是哪<span style="color: black;">有些</span>特征起到了关键的<span style="color: black;">功效</span>,使得它最后做出了<span style="color: black;">这般</span>的预测<span style="color: black;">或</span>决策。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">还有其他<span style="color: black;">各样</span>各样的方向。<span style="color: black;">咱们</span>自己<span style="color: black;">亦</span>在做<span style="color: black;">有些</span>,<span style="color: black;">期盼</span>能够把人工智能里面的<span style="color: black;">有些</span>黑盒子给打开,让<span style="color: black;">关联</span>的决策者能真正地理解AI在做什么,<span style="color: black;">最少</span><span style="color: black;">能够</span>去<span style="color: black;">检测</span>和验证人工智能找到的这些东西到底有用还是没用,再决定要不要<span style="color: black;">运用</span>它。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">当然<span style="color: black;">另一</span>一方面,我觉得AI for social good还<span style="color: black;">无</span>到瓶颈期,还有<span style="color: black;">非常多</span>问题值得去做,<span style="color: black;">因此</span><span style="color: black;">亦</span><span style="color: black;">期盼</span><span style="color: black;">更加多</span>的人能够关注这个方面的<span style="color: black;">科研</span>,能够更愿意去做<span style="color: black;">有些</span>这方面的<span style="color: black;">科研</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">当你说<span style="color: black;">非常多</span>问题<span style="color: black;">已然</span>被<span style="color: black;">处理</span>的时候,可能是图像<span style="color: black;">归类</span>、医疗图像识别<span style="color: black;">这般</span>的问题。这些问题做的人可能<span style="color: black;">已然</span><span style="color: black;">非常多</span>了,做了很久了,<span style="color: black;">已然</span>到了一个非常不错<span style="color: black;">然则</span>再进一步就很困难的<span style="color: black;">地区</span>。<span style="color: black;">然则</span></span><strong style="color: blue;"><span style="color: black;">我觉得AI for social good这块还是蓝海的状态,<span style="color: black;">由于</span>相对图像处理语音处理<span style="color: black;">或</span>自然语言处理,<span style="color: black;">这儿</span><span style="color: black;">无</span>一个规范的数据集或一个特定的问题,<span style="color: black;">大众</span>要<span style="color: black;">持续</span>去<span style="color: black;">研发</span>越来越好、越来越新的算法,要去<span style="color: black;">认识</span><span style="color: black;">区别</span>的问题,</span></strong><span style="color: black;">这个问题里面有什么是AI能够去<span style="color: black;">帮忙</span><span style="color: black;">处理</span>的,什么样的<span style="color: black;">办法</span>是最合适去<span style="color: black;">处理</span>的,最后怎么做系统性的测试,推动落地</span><strong style="color: blue;"><span style="color: black;">。</span></strong><span style="color: black;">除了算法设计以外,还有<span style="color: black;">非常多</span>其他的工作需要做。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:在<span style="color: black;">将来</span>2年内<span style="color: black;">或</span>5年内有什么想做的方向,<span style="color: black;">或</span>要进行的工作吗?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:<span style="color: black;">此刻</span>我<span style="color: black;">期盼</span>能够深挖的<span style="color: black;">行业</span>还是动物<span style="color: black;">守护</span>、<span style="color: black;">食品</span>救助和交通运输<span style="color: black;">关联</span>的<span style="color: black;">有些</span>问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">正在进行中的,一个是<span style="color: black;">咱们</span>正在跟世界自然基金会</span><span style="color: black;">(WWF)</span><span style="color: black;">合作动物<span style="color: black;">守护</span><span style="color: black;">关联</span>的新工作。<span style="color: black;">咱们</span>想<span style="color: black;">帮忙</span><span style="color: black;">她们</span>自动收集<span style="color: black;">资讯</span><span style="color: black;">报告</span>和政府报告,找到跟自然<span style="color: black;">守护</span>区和动物<span style="color: black;">守护</span><span style="color: black;">关联</span>的<span style="color: black;">文案</span>,<span style="color: black;">而后</span>把这些<span style="color: black;">文案</span>整理成可视化的、<span style="color: black;">能够</span>去直接查看和分析的形式,<span style="color: black;">这般</span>能够节省<span style="color: black;">她们</span>的时间。<span style="color: black;">她们</span><span style="color: black;">此刻</span>是靠人力在做这些事。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">这个工作<span style="color: black;">已然</span><span style="color: black;">起始</span>实地测试了,<span style="color: black;">已然</span>在WWF内部的系统里面用上了,<span style="color: black;">咱们</span>在<span style="color: black;">持续</span>收集反馈,改进工作。这个问题其实<span style="color: black;">无</span><span style="color: black;">触及</span>到博弈的问题,<span style="color: black;">触及</span>到<span style="color: black;">非常多</span>自然语言处理的东西。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">食品</span>救助那块,<span style="color: black;">咱们</span>还是<span style="color: black;">期盼</span>能够把现有的算法落地,<span style="color: black;">而后</span>去探寻新的问题。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">咱们</span>还有<span style="color: black;">有些</span>正在进行中的工作是跟网络安全<span style="color: black;">关联</span>的,这个问题里面很<span style="color: black;">显著</span>有博弈。<span style="color: black;">咱们</span>之前做过<span style="color: black;">有些</span>博弈论的模型和算法,<span style="color: black;">此刻</span>还是想导出更好的、更贴近<span style="color: black;">实质</span>的模型,更好的算法。<span style="color: black;">咱们</span><span style="color: black;">亦</span><span style="color: black;">期盼</span><span style="color: black;">咱们</span>做的东西更接近被<span style="color: black;">实质</span><span style="color: black;">运用</span>的<span style="color: black;">周期</span>。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:我有一个可能不是很切<span style="color: black;">实质</span>的想法,想到国内<span style="color: black;">近期</span>的疫情,在打疫苗上<span style="color: black;">大众</span><span style="color: black;">亦</span>是有<span style="color: black;">非常多</span>的犹豫。你的工作有可能<span style="color: black;">处理</span>鼓励<span style="color: black;">大众</span>打疫苗的问题吗?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:<span style="color: black;">倘若</span><span style="color: black;">已然</span>有<span style="color: black;">有些</span>奖励机制,<span style="color: black;">咱们</span><span style="color: black;">亦</span>许<span style="color: black;">能够</span>去分析这个机制,或许<span style="color: black;">能够</span>更精细化地奖励,让最后的效果更好,这是有可能的一个方向。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">给学习者的<span style="color: black;">意见</span></span></span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:<span style="color: black;">针对</span>想要来学博弈论跟深度学习结合的方向的学生,你会有给<span style="color: black;">她们</span>什么样的<span style="color: black;">意见</span>呢</span><span style="color: black;">?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:一个是<span style="color: black;">基本</span>还是要打的。本科<span style="color: black;">周期</span><span style="color: black;">倘若</span>有能够接触到博弈论<span style="color: black;">关联</span>的课程,以及深度学习、多智能体<span style="color: black;">关联</span>的课程,还是<span style="color: black;">意见</span>去学习一下,尤其是<span style="color: black;">倘若</span>有<span style="color: black;">有些</span>课程的项目,或是自己比较感兴趣的项目<span style="color: black;">能够</span>去做的话,有时候做项目的过程是能<span style="color: black;">帮忙</span>你更好地理解你学到的东西的。这个项目<span style="color: black;">乃至</span>不<span style="color: black;">必定</span>说你最后要发论文,可能<span style="color: black;">便是</span>去做<span style="color: black;">有些</span>对你<span style="color: black;">来讲</span>感兴趣的、有探索性的工作。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">知识分子</span></span></strong><span style="color: black;">:最后想问一下,你<span style="color: black;">得到</span>斯隆奖有什么感想吗?</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><span style="color: black;">方飞</span></span></strong><span style="color: black;">:我<span style="color: black;">亦</span>是真的<span style="color: black;">无</span>想到,<span style="color: black;">由于</span>我之前申请过一次,<span style="color: black;">无</span>拿到。今年又申请,我其实并<span style="color: black;">无</span>抱很高期望,后来拿到<span style="color: black;">亦</span>觉得非常荣幸。<span style="color: black;">由于</span>它不只是给计算机<span style="color: black;">行业</span>的,它是给<span style="color: black;">非常多</span><span style="color: black;">行业</span>的学者,<span style="color: black;">因此</span><span style="color: black;">好似</span>得到的关注度比较高。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">除了我自己获奖这件事让我很惊喜以外,<span style="color: black;">另一</span>一个我觉得挺惊喜的,是今年获奖者的名单里面有<span style="color: black;">非常多</span>的华人<span style="color: black;">或</span>华裔,还有<span style="color: black;">非常多</span>的女生,这个可能是之前我<span style="color: black;">无</span>看到的,说明<span style="color: black;">咱们</span></span><span style="color: black;">(华裔)</span><span style="color: black;">在北美做得挺好的。</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">参考资料:</span></strong></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"> V. Conitzer, T. Sandholm, Computing the Optimal Strategy to Commit to, EC’06, June 11–15, 2006, Ann Arbor, Michigan, USA</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">https://users.cs.duke.edu/~conitzer/commitEC06.pdf</span></span></p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">制版编辑</span><strong style="color: blue;"><span style="color: black;">|</span></strong><span style="color: black;">姜丝鸭</span></span></p>
你字句如珍珠,我珍藏这份情。 外贸论坛是我们的,责任是我们的,荣誉是我们的,成就是我们的,辉煌是我们的。 你的努力一定会被看见,相信自己,加油。 期待楼主的下一次分享!”
页:
[1]