# How to put AI ethics into practice? AI Human Impact **James Brusseau** · AI Policy Exchange · published 7 August 2020 *Author-accepted conservative reading edition, with documented limitations. Acceptance does not constitute whole-recording human acoustic verification.* ### Opening question ![Original video frame at 00:05.000](../frames/DISC-0064-VIS-001.jpg) **Displayed title-card wording:** > How to put AI ethics into practice? Question is how to put AI ethics into practice. Well, here's one way. We can make it possible for people to invest in technology companies where AI conforms to humans instead of having humans conform to the AI. ### Investment direction ![Original video frame at 00:20.000](../frames/DISC-0064-VIS-002.jpg) **Displayed presentation wording:** > AI HUMAN IMPACT INVESTING > Invest in those technology companies > where the AI conforms to humans, > instead of humans conforming to the AI And to do that, what we need is a model for objectively scoring AI-intensive businesses. I mean those companies that have artificial intelligence working at the core of their business operation. We need a way of scoring them on a humanitarian scale. ### Scoring companies ![Original video frame at 00:45.000](../frames/DISC-0064-VIS-003.jpg) **Displayed presentation wording:** > AI HUMAN IMPACT INVESTING > Objectively score AI-intensive companies > on an humanitarian scale And the good news is we have a lot of help. There've been more than 80 principles produced in the last few years, and more are coming, along with principles of the principles and principles of principles of the principles, and so on. So we have a lot of sets of ideas that try to define what kinds of requirements are necessary for artificial intelligence to serve the human condition. ### Displayed research quotation ![Original video frame at 00:55.000](../frames/DISC-0064-VIS-004.jpg) **Displayed third-party research extract — Thomas Ploug and Søren Holm; Elsevier branding.** Visible title: “The four dimensions of contestable AI diagnostics- A patient-[partly obscured by the AI Policy Exchange overlay] / approach to explainable AI.” > In response to these problems, public committees and expert groups, research institutions and private companies have in recent years issued reports on and guidelines for responsible use of AI. A systematic review of the corpus of such guidelines found 84 documents containing ethical principles and guidelines for the use of AI [11]. The review identifies a significant global convergence on the importance of, among others, the transparency or explainability of AI decision-making. The review also finds, however, a significant variance in the posited features of transparency. Thus, for instance, transparency is posited to 1) serve very *Editorial note: the extract continues beyond the bottom of the recorded frame. The line-wrap division “trans-parency” is joined above; no missing continuation is reconstructed.* **AI-generated visual description:** Yellow highlighting marks the finding about 84 documents containing ethical principles and guidelines for AI. The spoken “more than 80 principles” is retained separately above; the displayed text refers to documents. ### Build the framework ![Original video frame at 01:20.000](../frames/DISC-0064-VIS-005.jpg) **Displayed presentation wording:** > AI HUMAN IMPACT INVESTING > Build the framework Now, we need to boil down to key elements, and this is what that model is going to look like in the end. I'm fast-forwarding. There are three pillars of AI ethics: the personal, the social and the technical. And each one of those supports three values. So, as you can see here, within the personal we have autonomy, dignity and privacy. Within the social we have fairness, solidarity and sustainability. And then on the technical front we have performance, safety and accountability. ### Values and key performance indicators ![Original video frame at 01:30.000](../frames/DISC-0064-VIS-006.jpg) **Displayed presentation wording and table:** > VALUES / KEY PERFORMANCE INDICATORS | Personal Freedom | Social Wellbeing | Technical Trustworthiness | |---|---|---| | Autonomy | Fairness | Performance | | Dignity | Solidarity | Safety | | Privacy | Sustainability | Accountability | And now, one quick point: the critical category is all the values of personal freedom, that is, autonomy, dignity and privacy. These are the key performance indicators for AI ethical evaluations of companies, because the purpose of humanitarian artificial intelligence is to increase individual human potential. The reason we have the ethics of AI in the first place is to help each one of us as individuals do more independently. ### Apply the model ![Original video frame at 02:35.000](../frames/DISC-0064-VIS-007.jpg) **Displayed presentation wording:** > AI HUMAN IMPACT INVESTING > Apply the model So that's the model. We have a set of AI humanitarian values, and then we can go through and apply them as grades. For example, we might say that a company like Apple, that does well in amplifying personal autonomy and also does well with privacy protection, we might say a company like that scores highly. ### Apple example ![Original video frame at 02:50.000](../frames/DISC-0064-VIS-008.jpg) **Displayed table wording:** | Personal Freedom | Social Wellbeing | Technical Trustworthiness | |---|---|---| | Autonomy | Fairness | Performance | | Dignity | Solidarity | Safety | | Privacy | Sustainability | Accountability | **AI-generated visual description:** A grey Apple logo appears above the table. Green strokes underline Autonomy and Privacy in the Personal Freedom column; Dignity is not underlined. No numerical grades are displayed. Whereas, by contrast, a company like TikTok, which is part of ByteDance, a company or a conglomerate like that captures users in addictive dark patterns, dopamine patterns, let's say. So they're going to fail the test of increasing self-determination. They're going to fail the test of autonomy, and they're also going to fail the test of privacy, as they capture all sorts of data without any acknowledgement and spread it to who knows what places. ### TikTok and addictive use ![Original video frame at 03:00.000](../frames/DISC-0064-VIS-009.jpg) **Displayed third-party quotation — Forbes, John Koetsier, Senior Contributor; Consumer Tech.** The screenshot shows 18 January 2020, 02:04pm EST. > Digital Crack Cocaine: The Science Behind TikTok’s Success > TikTok gained 682 million new users last year, each of whom spend an average of 50 minutes a day in the app. That’s mind-blowing, hockey-stick, off-the-charts growth that we haven’t seen since Pokemon Go. > > But what makes TikTok so successful? Displayed author note: “John Koetsier is a journalist, analyst, author, and speaker.” **AI-generated visual description:** The article screenshot includes four vertical TikTok video thumbnails below the text. Small thumbnail overlays are not reliably transcribed; they are retained in the frame. ### Privacy example added to the article comparison ![Original video frame at 03:15.000](../frames/DISC-0064-VIS-010.jpg) **Left — repeated displayed third-party quotation:** **Displayed third-party quotation — Forbes, John Koetsier, Senior Contributor; Consumer Tech.** The screenshot shows 18 January 2020, 02:04pm EST. > Digital Crack Cocaine: The Science Behind TikTok’s Success > TikTok gained 682 million new users last year, each of whom spend an average of 50 minutes a day in the app. That’s mind-blowing, hockey-stick, off-the-charts growth that we haven’t seen since Pokemon Go. > > But what makes TikTok so successful? Displayed author note: “John Koetsier is a journalist, analyst, author, and speaker.” **Right — displayed third-party quotation, Forbes, Zak Doffman, Contributor; Cybersecurity.** The screenshot shows 26 June 2020, 01:46am EDT. > Warning—Apple Suddenly Catches TikTok Secretly Spying On Millions Of iPhone Users Displayed author note: “I write about security and surveillance.” **AI-generated visual description:** A second article screenshot is added to the right of the Koetsier screenshot. Its photograph shows the TikTok mark and wordmark; the displayed credit is “AFP VIA GETTY IMAGES.” This meaningful build preserves the left article as a repeated occurrence. ### Apple and TikTok comparison ![Original video frame at 03:25.000](../frames/DISC-0064-VIS-011.jpg) **Displayed presentation wording:** > AI HUMAN IMPACT **AI-generated visual description:** An Apple-sign photograph sits above a TikTok app-store screenshot, with a vertical double-headed arrow alongside the two. There is no numeric scale. Photographer credits are not displayed here. **Displayed third-party app-store wording:** > tik tok > TikTok - Make Your Day > Real People. Real Videos. > GET > Ad > In-App Purchases > TikTok is THE destination for mobile videos. On TikTok, short-form videos are exciting, spontaneous, and genuine. Whether *Editorial note: the description ends at “Whether” in the screenshot; its continuation is not supplied.* So we have a company that scores well, like Apple, a company that scores poorly, TikTok. And then that's the way this model would develop. We find each company and define it in terms of each of these nine values, that is, score each company in terms of its autonomy, in terms of its dignity, and down the list to accountability and safety and so on. And in that way, we could form a hierarchy of humanitarian AI companies, from better to worse. ### Comparison with the nine values ![Original video frame at 03:55.000](../frames/DISC-0064-VIS-012.jpg) **Displayed presentation wording:** > AI HUMAN IMPACT > Autonomy > Dignity > Privacy > Fairness > Solidarity > Sustainability > Performance > Safety > Accountability **AI-generated visual description:** The Apple photograph, TikTok app-store screenshot and vertical double-headed arrow reappear; the nine-value list is now added at the left. This is a meaningful build relative to the earlier comparison. **Displayed third-party app-store wording:** > tik tok > TikTok - Make Your Day > Real People. Real Videos. > GET > Ad > In-App Purchases > TikTok is THE destination for mobile videos. On TikTok, short-form videos are exciting, spontaneous, and genuine. Whether *Editorial note: the description ends at “Whether” in the screenshot; its continuation is not supplied.* So where we end up is with a set of what we could call investment-grade AI humanitarian companies. And then that leads back to the original question we had, which was: how to put AI ethics into practice. And one practical answer is to develop and participate in this kind of AI human impact investing. And then, one last point: it is your money. This kind of investment strategy, this is not charity, it's not altruism. This is a way to benefit financially, because I believe that AI-intensive companies, companies that ensure the technology genuinely serves human being, will outperform its competitors in the medium to long term. ### Investment framework reminder ![Original video frame at 04:30.000](../frames/DISC-0064-VIS-013.jpg) **Displayed presentation wording:** > AI HUMAN IMPACT INVESTING > Autonomy > Dignity > Privacy > Fairness > Solidarity > Sustainability > Performance > Safety > Accountability **AI-generated visual description:** The nine values are displayed as a single list. The earlier Apple and TikTok images and arrow are absent. So there's that too.