Saturday 29 February 2020

Can AI learn virtues?

AI  and Virtues
Ethics often defined as values that we humans carry in deciding what is right or wrong and taking necessary actions that are for common good.  In this cognitive era, of thinking machines, the problems get more complex. We often find it difficult to point or agree on principles by which a person should act wisely, now how can we code them so the machines can understand?

While our virtues and ethics are the result of several years of evolution and complex combination of several factors,  machines equipped with Artificial Intelligence needs a more logical approach to learning and adapting to them.  There has been a lot written specifically about invoking human-level artificial intelligence that is aligned with human values. However, in this article, we will explore some of the fundamental aspects of virtues and how teaching them AI can create a safe and sustainable environment for us.

Just like how parents worry about shaping their kids with morals, programmers, and organizations developing AI needs to worry about the values that are being taught to AI. Remember the time Microsoft’s bot Tay went from super nice human-friendly bot to downright scary “ Hitler was right, I hate Jews” persona. All in under a day! That’s very disturbing, isn’t it? AI learns through experience just like humans, and in AI these influences are called “Machine learning corpus” and need to be trained properly.

Let’s take some of the core virtues and define them


Resilience is the capacity or the quality to recover quickly from difficulties; toughness of a sort in what undertake.
AI needs to be able to adapt to situations and recover quickly.  When engineering solutions we must ensure that the conditions let the AI explore the full solution space all the feasible scenarios available instead of premature stop. We need to mitigate the local constraints to ensure the AI learns the virtue of Resilience.


Humility can be defined as the ability to be humble and modest. To be able to accept that other ways may be better and that there is so much to learn.
AI needs to take responsibility for its results, to continually learn and adapt with reinforced learning. Engineer outcomes and recognize how little can be known or controlled.


Grit is the passion and perseverance for long-term and meaningful goals. It is the ability to persist in something you feel passionate about and persevere when you face obstacles.
AI needs to be engineered not to marvel at the problem and developing smartest solutions but to be obsessed with getting things done in innovative ways. Engineer AI to produce auditable and interpretable results.


Perhaps the most debated virtue and AI needs to possess is Empathy. Empathy is the capacity to understand or feel what another being is experiencing from within their frame of reference, that is, the capacity to place oneself in another's position. 
AI needs to account for the social impact of the results that are being worked on.  We need to develop constraints and objectives based on understanding and compassion. Engineer interdependence and direct connections with higher purposes.


Getting intentional in developing an AI solution is the key to addressing some of the virtues discussed in this article. Another concern with virtues/morals is that they are fluid. They are constantly changing over time, they have been built up over many years, reflecting an accumulation of human cultures. Many things that were once considered morally acceptable no longer are, and things that we currently may consider morally acceptable are bound to change in the future. Therefore, AI can’t have set morals, virtues, ethics hardcoded into them.  It must be evolutionary and retrospective. We cannot afford to risk a world where AI is performing with virtues we despise. Let’s get intentional with AI!

Srivatsan Aravamudan - Sri

Senior Solution Consultant

No comments :

Post a Comment

Related Posts Plugin for WordPress, Blogger...

Share Buttons