{"id":31602,"date":"2022-10-19T09:53:00","date_gmt":"2022-10-19T00:53:00","guid":{"rendered":"https:\/\/smallake.kr\/?p=31602"},"modified":"2022-10-19T09:53:23","modified_gmt":"2022-10-19T00:53:23","slug":"ai%eb%a5%bc-%eb%84%98%ec%96%b4%ec%84%9c-xai%eb%a1%9c-ai%ea%b0%80-%eb%8b%b5%ed%95%b4%ec%95%bc-%ed%95%98%eb%8a%94-%ec%a7%88%eb%ac%b8%eb%93%a4","status":"publish","type":"post","link":"https:\/\/smallake.kr\/?p=31602","title":{"rendered":"AI\ub97c \ub118\uc5b4\uc11c XAI\ub85c… AI\uac00 \ub2f5\ud574\uc57c \ud558\ub294 \uc9c8\ubb38\ub4e4"},"content":{"rendered":"

1.
\n \uc624\ub798 \uc804 \uc54c\ud30c\uace0\uc640 \uc774\uc138\ub458 9\ub2e8\uc758 \ub300\uad6d\uc744 \uc911\uacc4\ud558\uc600\uc2b5\ub2c8\ub2e4. \uc54c\ud30c\uace0\uac00 \ubc14\ub451\uc54c\uc744 \ub193\uc73c\uba74 \ud574\uc124\uc744 \ud558\uc5ec\uc57c \ud558\ub294\ub370 \uacbd\ud5d8\uc0c1 \ubcf4\uc9c0 \ubabb\ud55c \uc218\uc77c \uacbd\uc6b0 \ud574\uc11d\uc744 \ud558\uc9c0 \ubabb\ud558\ub294 \uacbd\uc6b0\uac00 \ub9ce\uc558\uc2b5\ub2c8\ub2e4. \uc5b4\ub5a4 \uacbd\uc6b0\ub124\ub294 \ub300\uad6d\uc774 \uc911\ubc18\uc744 \ub118\uc5b4\uc11c\uc11c \ucd08\ubc18\uc5d0 \ub450\uc5c8\ub358 \uc218\uc758 \uc758\ubbf8\ub97c \uc774\ud574\ud558\ub294 \uacbd\uc6b0\ub3c4 \uc788\uc5c8\uc2b5\ub2c8\ub2e4.\ubc14\ub451\uc744 \ubcf4\ub294 \uc0ac\ub78c\uc740 \ub2f9\uc5f0\ud788 \uc9c8\ubb38\uc744 \ud569\ub2c8\ub2e4.<\/p>\n

“\uc65c \uc774 \uc218\ub97c \ub450\uc5c8\uc744\uae4c?”<\/p><\/blockquote>\n

\ubcf4\ud1b5 \ud504\ub85c\uae30\uc0ac\uc758 \ub300\uad6d\uc740 \ub300\uad6d\uc744 \ub9c8\uce5c \ud6c4 \uc124\uba85\uc744 \ud569\ub2c8\ub2e4. \uc774\ub7f4 \uacbd\uc6b0\uc640 \ube44\uad50\ud558\uc5ec \uc9c8\ubb38\uc774 \uac00\ub2a5\ud569\ub2c8\ub2e4.<\/p>\n

“\uc54c\ud30c\uace0\ub294 \ub300\uad6d\uc744 \ud574\uc124\ud560 \uc218 \uc788\uc744\uae4c?”<\/p><\/blockquote>\n

\uc800\ub294 \uc54c\ud30c\uace0\uac00 \ud574\uc124\ud560 \uc218 \uc5c6\uc744 \uac83\uc774\ub77c\uace0 \uc0dd\uac01\ud588\uc2b5\ub2c8\ub2e4. \uadf8\ub7ec\ub098 2016\ub144 \uc54c\ud30c\uace0\ub294 \uc774\uc138\ub3cc\uc758 \uc218\ub97c \uc774\ub807\uac8c \ubcf4\uc558\ub2e4 -1<\/a>\uc744 \ubcf4\uba74 \uc54c\ud30c\uace0\ub294 \uc54c\ud30c\uace0\ub098\ub984\uc758 \ud574\uc124\uc744 \ud558\uace0 \uc788\uc5c8\uc2b5\ub2c8\ub2e4. \uc54c\ud30c\uace0\uac00 \ub9d0\ub85c \ud45c\ud604\ud558\uc9c0 \uc54a\uc558\uc9c0\ub9cc \uc758\uc0ac\uacb0\uc815\ud560 \ub54c \uc608\uc0c1\ud588\ub358 \ub3cc\uc758 \ud750\ub984=\ucc38\uace0\ub3c4\ub97c \uac00\uc9c0\uace0 \uc788\uc5c8\uace0 \uc774\ub97c \uacf5\uac1c\ud588\uc2b5\ub2c8\ub2e4. \uc717 \uadf8\ub9bc\uc758 1\uc774 \uc54c\ud30c\uace0\uac00 \ub193\uc740 \uc218\uc778\ub370 \uc774\ub97c \uacb0\uc815\ud560 \ub54c \uc608\uc0c1\ud588\ub358 \uc218\uc21c\uc774\ub77c\uace0 \ud569\ub2c8\ub2e4.<\/p>\n

\"\"<\/a>
\n\"\"<\/a><\/p>\n

\ub2e4\uc2dc \ucc98\uc74c \uc9c8\ubb38\uc73c\ub85c \ub3cc\uc544\uac00\uc11c \uc54c\ud30c\uace0\ub294 \uc54c\ud30c\uace0 \ub098\ub984\uc73c\ub85c \ub300\uad6d\uc744 \ud574\uc124\ud569\ub2c8\ub2e4. \ubb3c\ub860 \uc6b0\ub9ac\uac00 \uc544\ub294 \ud574\uc124\uc740 \uc544\ub2c8\uc9c0\ub9cc…<\/p>\n

\ub610\ub2e4\ub978 \uc9c8\ubb38\uc744 \ud574\ubcf4\uc8e0. \uc0ac\ud68c\uc5d0\uc11c \uc790\uc8fc \ub4e3\ub294 \uc9c8\ubb38\uc785\ub2c8\ub2e4.<\/p>\n

“\ub2e4\uc74c \ub274\uc2a4\uc11c\ube44\uc2a4\uc758 \uc54c\uace0\ub9ac\uc998\uc740 \uc5b4\ub5a4 \uae30\uc900\uc73c\ub85c \ucd94\ucc9c\uc744 \ud560\uae4c?”
\n“\ubc30\ubbfc, \ucfe0\ud321\uc774\uce20, \uc694\uae30\uc694 \ubc30\ub2ec \ud50c\ub7ab\ud3fc 3\uc0ac\ub294 \uc5b4\ub5a4 \uae30\uc900\uc73c\ub85c \ubc30\ub2ec\ub178\ub3d9\uc790\uc5d0\uac8c \uc791\uc5c5\uc744 \ubc30\ub2f9\ud560\uae4c?”<\/p><\/blockquote>\n

\uc774\uc640 \uad00\ub828\ud558\uc5ec \uc804\ubb38\uac00\uac00 \uc124\uacc4 \ubc0f \uc18c\uc2a4\ucf54\ub4dc\ub97c \uac80\uc99d\ud558\ub294 \ubc29\uc2dd\uc73c\ub85c \uc870\uc0ac\ud558\ub294 \uacbd\uc6b0\ub294 \uc788\uc9c0\ub9cc \uc774\ud574\ub2f9\uc0ac\uc790 \ud639\uc740 \uc18c\ube44\uc790\uac00 \uad00\ub828\ud558\uc5ec \ud655\uc778\ud560 \uc218 \uc788\ub294 \ubc29\ubc95\uc740 \uc5c6\uc2b5\ub2c8\ub2e4. \ub610\ub2e4\ub978 \uc9c8\ubb38\ub3c4 \uac00\ub2a5\ud569\ub2c8\ub2e4.<\/p>\n

“\ub85c\ubcf4\ub4dc\ubc14\uc774\uc800\ub294 \ubb34\uc2a8 \uae30\uc900\uc73c\ub85c \uc885\ubaa9\uc744 \uc120\uc815\ud560\uae4c? \uadf8\ub9ac\uace0 \ubb34\uc2a8 \uc774\uc720\ub85c \uc7ac\uc120\uc815\ud560\uae4c?”<\/p><\/blockquote>\n

\uc778\uacf5\uc9c0\ub2a5\uc740 \uc218\ub9ce\uc740 \ub370\uc774\ud0c0\ub97c \ud1b5\ud558\uc5ec \ud559\uc2b5\ud558\uace0 \ud559\uc2b5\ud55c \uacb0\uacfc\uc5d0 \ub530\ub77c \uc758\uc0ac\uacb0\uc815\uc744 \ud569\ub2c8\ub2e4. \ucc98\uc74c \uc0ac\ud68c\ub294 \uc758\uc0ac\uacb0\uc815\uc5d0\ub9cc \uc9d1\uc911\ud558\uc600\uc2b5\ub2c8\ub2e4. \uc0ac\ud68c\ubb38\uc81c\uac00 \ub418\uc9c0 \uc54a\ub294\ub2e4\uba74 \uacfc\uc815 \uc790\uccb4\uc5d0 \uad00\uc2ec\uc744 \ub450\uc9c0 \uc54a\uc558\uc2b5\ub2c8\ub2e4. \uadf8\ub7ec\ub098 \uc778\uacf5\uc9c0\ub2a5\uc774 \ubcf4\ud3b8\ud654\ud558\uace0 \uc601\ud5a5\ub825\uc774 \ucee4\uc9c0\uba74\uc11c \uad00\uc2ec\uc774 \ub354 \ub113\uc5b4\uc9d1\ub2c8\ub2e4. <\/p>\n

“\uc5b4\ub5a4 \uadfc\uac70\ub85c \uc774\ub7f0 \uc758\uc0ac\uacb0\uc815\uc744 \ud588\ub098?”<\/p><\/blockquote>\n

\uc0ac\ub78c\uacfc \uc0ac\ub78c\uc0ac\uc774\uc5d0\uc11c\ub294 \ub2f9\uc5f0\ud55c \uc9c8\ubb38\uc785\ub2c8\ub2e4. \ub2f9\uc5f0\ud55c \uc9c8\ubb38\uc744 \uc778\uacf5\uc9c0\ub2a5\uc5d0 \ud558\ub294 \ub2e8\uacc4\uc5d0 \uc654\uc2b5\ub2c8\ub2e4. \uc815\ud655\ud788\ub294 \uc778\uacf5\uc9c0\ub2a5\uc11c\ube44\uc2a4\ub97c \ud558\ub294 \uac1c\uc778 \ud639\uc740 \uc9d1\ub2e8\uc5d0 \ud569\ub2c8\ub2e4. \uc774\uac83\uc774 \uc124\uba85\uac00\ub2a5\ud55c \uc778\uacf5\uc9c0\ub2a5(Explainable AI)\uc785\ub2c8\ub2e4. \uae08\uc735\uc5d0\uc11c\ub3c4 XAI\ub294 \ud654\ub450\uc785\ub2c8\ub2e4. \uc774\ubbf8 \uae08\uc735\uc704\uc758 \uae08\uc735AI \uac1c\ubc1c\ud65c\uc6a9 \uc548\ub0b4\uc11c<\/a>\uc5d0\uc11c \uc18c\uac1c\ud558\uc600\uc9c0\ub9cc \uae08\uc735\uc704\uc6d0\ud68c\ub294 \uc774\ubbf8 \uc124\uba85\uac00\ub2a5\ud55c AI\ub97c \uae30\uc900\uc73c\ub85c \uc81c\uc2dc\ud558\uc600\uc2b5\ub2c8.<\/p>\n

\"\"<\/p>\n

2.
\n \ud558\ubc14\ub4dc\ube44\uc9c0\ub2c8\uc2a4\ub9ac\ubdf0 When \u2014 and Why \u2014 You Should Explain How Your AI Works <\/a>\uc744 “\uc65c, \uc5b8\uc81c” \uc124\uba85\uac00\ub2a5\ud55c AI\uac00 \ud544\uc694\ud55c\uc9c0\ub97c \ubcf4\uc5ec\uc90d\ub2c8\ub2e4.<\/p>\n

What Makes an Explanation Good?<\/strong><\/p>\n

A good explanation should be intelligible to its intended audience, and it should be useful, in the sense that it helps that audience achieve their goals. When it comes to explainable AI, there are a variety of stakeholders that might need to understand how an AI made a decision: regulators, end-users, data scientists, executives charged with protecting the organization\u2019s brand, and impacted consumers, to name a few. All of these groups have different skill sets, knowledge, and goals \u2014 an average citizen wouldn\u2019t likely understand a report intended for data scientists.<\/p>\n

So, what counts as a good explanation depends on which stakeholders it\u2019s aimed at. Different audiences often require different explanations.<\/p>\n

For instance, a consumer turned down by a bank for a mortgage would likely want to understand why they were denied so they can make changes in their lives in order to get a better decision next time. A doctor would want to understand why the prediction about the patient\u2019s illness was generated so they can determine whether the AI notices a pattern they do not or if the AI might be mistaken. Executives would want explanations that put them in a position to understand the ethical and reputational risks associated with the AI so they can create appropriate risk mitigation strategies or decide to make changes to their go to market strategy.<\/p>\n

Tailoring an explanation to the audience and case at hand is easier said than done, however. It typically involves hard tradeoffs between accuracy and explainability. In general, reducing the complexity of the patterns an AI identifies makes it easier to understand how it produces the outputs it does. But, all else being equal, turning down the complexity can also mean turning down the accuracy \u2014 and thus the utility \u2014 of the AI. While data scientists have tools that offer insights into how different variables may be shaping outputs, these only offer a best guess as to what\u2019s going on inside the model, and are generally too technical for consumers, citizens, regulators, and executives to use them in making decisions.<\/p>\n

Organizations should resolve this tension, or at least address it, in their approach to AI, including in their policies, design, and development of models they design in-hour or procure from third-party vendors. To do this, they should pay close attention to when explainability is a need to have versus a nice to have versus completely unnecessary.
\nWhen We Need Explainability<\/strong><\/p>\n

Attempting to explain how an AI creates its outputs takes time and resources; it isn\u2019t free. This means it\u2019s worthwhile to assess whether explainable outputs are needed in the first place for any particular use case. For instance, image recognition AI may be used to help clients tag photos of their dogs when they upload their photos to the cloud. In that case, accuracy may matter a great deal, but exactly how the model does it may not matter so much. Or take an AI that predicts when the shipment of screws will arrive at the toy factory; there may be no great need for explainability there. More generally, a good rule of thumb is that explainability is probably not a need-to-have when low risk predictions are made about entities that aren\u2019t people. (There are exceptions, however, as when optimizing routes for the subway leads to giving greater access to that resource to some subpopulations than others).<\/p>\n

The corollary is that explainability may matter a great deal, especially when the outputs directly bear on how people are treated. There are at least four kinds of cases to consider in this regard.
\nWhen regulatory compliance calls for it.<\/strong><\/p>\n

Someone denied a loan or a mortgage deserves an explanation as to why they were denied. Not only do they deserve that explanation as a matter of respect \u2014 simply saying \u201cno\u201d to an applicant and then ignoring requests for an explanation is disrespectful \u2014 but it\u2019s also required by regulations. Financial services companies, which already require explanations for their non-AI models, will plausibly have to extend that requirement to AI models, as current and pending regulations, particularly out of the European Union, indicate.
\nWhen explainability is important so that end users can see how best to use the tool.<\/strong><\/p>\n

We don\u2019t need to know how the engine of a car works in order to drive it. But in some cases, knowing how a model works is imperative for its effective use. For instance, an AI that flags potential cases of fraud may be used by a fraud detection agent. If they do not know why the AI flagged the transaction, they won\u2019t know where to begin their investigation, resulting in a highly inefficient process. On the other hand, if the AI not only flags transactions as warranting further investigation but also comes with an explanation as to why the transaction was flagged, then the agent can do their work more efficiently and effectively.
\nWhen explainability could improve the system.<\/strong><\/p>\n

In some cases, data scientists can improve the accuracy of their models against relevant benchmarks by making tweaks to how it\u2019s trained or how it operates without having a deep understanding of how it works. This is the case with image recognition AI, for example. In other cases, knowing how the system works can help in debugging AI software and making other kinds of improvements. In those cases, devoting resources to explainability can be essential for the long-term business value of the model.
\nWhen explainability can help assess fairness.<\/p>\n

Explainability comes, broadly, in two forms: global and local. Local explanations articulate why this particular input led to this particular output, for instance, why this particular person was denied a job interview. Global explanations articulate more generally how the model transforms inputs to outputs. Put differently, they articulate the rules of the model or the rules of the game. For example, people who have this kind of medical history with these kinds of blood test results get this kind of diagnosis.<\/p>\n

In a wide variety of cases, we need to ask whether the outputs are fair: should this person really have been denied an interview or did we unfairly assess the candidate? Even more importantly, when we\u2019re asking someone to play by the rules of the hiring\/mortgage lending\/ad-receiving game, we need to assess whether the rules of the game are fair, reasonable, and generally ethically acceptable. Explanations, especially of the global variety, are thus important when we want or need to ethically assess the rules of the game; explanations enable us to see whether the rules are justified.<\/p><\/blockquote>\n

JP Morgan\uc758 AI \ucc45\uc784\uc790\ub294 \uc778\ud130\ubdf0\uc5d0\uc11c XAI\uac00 \ud544\uc694\ud55c \uc774\uc720\ub97c \uc544\ub798\uc640 \uac19\uc774 \uc81c\uc2dc\ud569\ub2c8\ub2e4. 2019\ub144 \uc778\ud130\ubdf0\uc785\ub2c8\ub2e4.<\/p>\n

Knowledge at Wharton: How is AI reshaping the financial services industry?<\/p>\n

Apoorv Saxena: AI is impacting every industry. AI is making a substantial, wide-ranging impact because it is used with data, and every industry is increasingly becoming data-driven. Companies across every industry are looking to gather and use more data. They want to better understand who their customers are, how they interact with them, the services they provide, and how they can improve those services and experiences. Every activity is becoming data-driven.<\/p>\n

AI is allowing companies like Google, Facebook and Amazon to achieve hyper-scale. You can get personalized news feeds in real-time. A grocery store or a bookstore like Amazon can serve hundreds of millions of users globally. That is possible when you inject AI into every piece of your business process. Now, transfer this to AI and finance. The future of AI in finance is a bank that can serve billions of people and provide personalized services.
\n<\/br>
\nKnowledge at Wharton: What are some of the opportunities and challenges in implementing this vision?
\n<\/br>
\nSaxena: The opportunity is that AI will let banks provide services in much more personalized, highly scalable and customized ways. The challenges include the ability to explain your AI \u2013 what we call \u201cAI explainability.\u201d When AI is used, the regulatory environment requires banks to justify or rationalize decisions. JPMorgan is trying to be the leader in applying \u201cexplainability\u201d to financial markets. Another challenge is to ensure confidentiality, since a lot of the data in finance is personal information or highly confidential.<\/strong>
\n<\/br>
\nKnowledge at Wharton: If you look at financial institutions, technology companies and telecom companies \u2014 which are all broadly involved in mobile money and offering financial services to a massive number of customers \u2014 who do you think is best positioned to win in AI and why?<\/p>\n

Saxena: The essence of finance and banking \u2013 banking in particular \u2013 is trust. User trust is key. The person on the other side wants to trust you with their most valuable assets, and with their most valuable information. And they want you to manage these assets in a way that is compliant with regulations.<\/p>\n

The second factor is customer service. Customers are looking for you to provide the best service possible, in a manner that conforms to the trust. If you break it down to fundamentals, finance is a service built around trust and regulation.<\/p>\n

Anybody who can replicate that model of trust, regulatory compliance and client service is well-positioned to be a player in this space. It does require very deep domain knowledge. There are some areas of banking, like payments, which involve highly skilled operations, but which are not deep-domain. Many other financial services [require] extremely detailed and very deep domain understanding. For example, how do you manage M&As? How do you create complex securities? These are non-trivial and highly domain-specific, and there will be space for banks to continue to provide those services, given their expertise, existing client relationships and thorough understanding of the complex environment.
\nWhat\u2019s Behind JPMorgan Chase\u2019s Big Bet on Artificial Intelligence? <\/a>\uc911\uc5d0\uc11c<\/p><\/blockquote>\n

\uc774\ub97c \uc704\ud574\uc11c JP Morgan\uc758 AI \uc5f0\uad6c\uc18c\ub294 Seven challenges for harmonizing explainability requirements<\/a>\ub77c\ub294 \uc81c\ubaa9\uc758 \ub17c\ubb38\uc744 \ubc1c\ud45c\ud569\ub2c8\ub2e4. XAI \uc5f0\uad6c\ub3d9\ud5a5\uc744 \uc0b4\ud53c\uace0 \uc77c\uacf1\uac1c\uc758 \ub3c4\uc804\uacfc\uc81c\ub97c \uc120\uc815\ud558\uc600\uc2b5\ub2c8\ub2e4.<\/p>\n

\nChallenge 1. The intent and scope of explanations matter
\nChallenge 2. The type of data and type of model matter\\
\nChallenge 3. The human factors around intent and scope of explanations matter.
\nChallenge 4. There is no consensus around evaluating the correctness of explanations.
\nChallenge 5. XAI techniques in general lack robustness and have strong basis dependence
\nChallenge 6. Feature importance explanation methods can be manipulated
\nChallenge 7. Too detailed an explanation can compromise a proprietary model that was intended to be kept confidential.<\/p><\/blockquote>\n