Where is AI taking us?

 29 March 2021

While I believe in the positives of the ongoing AI revolution in our lives, I got some contrary (and highly relevant) view-points from a conversation between the Nobel laureates Kazuo Ishiguro and Venki Ramakrishnan on the impact of AI on our civilization.  A quick summary:

* Inbuilt Biases : AI takes decisions based on large data sets which come with inbuilt existing biases (e.g gender, race).  Once acceptable biases e.g. about slavery if they are managed by humans can be changed over a period of time. But when these biases are inbuilt into an algorithm - and are opaque - it will be very intimidating to challenge and question, and will be very difficult to over-turn. AI will become a kind of Delphi oracle. The most worrying aspect : it is not clear how these existing biases can be removed.

* Deep Fakes created through AI is going to be an increasing problem - and will make it difficult for humans to differentiate truth from fiction.  The only solution will be to counter this with an AI that will detect Deep Fakes i.e. AI vs AI. This concept of AI vs AI is difficult to understand, and may leave humans totally bewildered as to what AI to believe in, and whether we can at all believe what AI tells us if Deep Fakes proliferate.

* Impact on jobs : While we console ourselves saying that job losses due to AI will be made up by job creation in other areas - the fact is new types of job creation takes decades to materialise.  In the 1st industrial revolution, it took over 100 years before the average worker was better off. In the meantime their lives were disrupted, and they were thrown out of work and into poverty.  The current AI revolution will create a few high priests who are still needed, and the rest made redundant - and we are already seeing signs of this as once flourishing occupations are disappearing and the once successful practioners take on less remunerative tasks to earn a livelihood. 

* DNA Tinkering : One of the biggest impact of AI is on gene editing.  While this has a lot of positives, it can also (like plastic surgery) be used by the privileged to create "superior characteristics" - whether that is intellectual, cognitive, or physical. It will change the whole idea of meritocracry on which our society is based - as it will be possible for the privileged to buy their way into meritocracy.  Rules will need to be set to prevent this - and will have to be implemented globally to prevent certain countries exploiting this - a difficult challenge.

* Monitoring AI : Humans have already lost the ability to control - and let alone fully understand - the pace at which algorithms are governing our lives.  Having a human to monitor AI and ensure it stays on the right path will be like "having a retired watchman supervise a stadium full of rioting football fans".  What happens then to Asimov's rules for robots (not doing harm etc) - can they be sustained?

* Manipulating Emotions : We pretend we are rational creatures but, deep down, we are actually highly emotional beings.  And we assume that our biggest advantage over AI is our monopoly on understanding emotions and asking the right questions. But once AI starts understanding human emotions, it will be able to detect the emotions that are there in society - the anger, the frustration, the hopes.  It will then be able to generate empathy and trigger emotions among humans - make them laugh or cry - and therefore can even manage a political campaign manipulating these emotions.

The biggest worry about AI is not whether AI can write a great novel or poetry.  The biggest worry is when AI starts writing our Constitutions.  AI could come up with the next big idea - democracy or Nazism.  As AI will be perceived as "neutral" and independent of biases, it will be far easier for AI to manipulate human emotions than for humans to manipulate human emotions .  And once it understands how to manipulate human emotions we have got much bigger things to be concerned about.