Decide for Yourself

Since the enlightenment we try to overcome the "self-inflicted immaturity". Immaturity is the lack of using our mental capacity without help. This immaturity is self-inflicted if it's not for lack of wits but for lack courage and determination. In other words: when are cowards or lazy.

Decision making is risky. We might be wrong and embarrass ourselves. We might have to justify our decision.

There are ways to reduce the risk:

• Do it like everybody else
• Do it like it has always been done
• Let the computer decide

Hm, the last point is particularly interesting. Why should a computer be a reliable agent? Because our experience with computers has been that they are impartial, deterministic, precise, data-based and that they provide repeatable results that can be fully explained.

They might have been annoying at times and not part of our real life but within their constraints, they have been pretty trustworthy.

This is no longer true in the AI world.

No Longer Impartial

What I always liked about computers was their impartial treatment of us programmers. When the code was wrong, the compiler failed, no matter if the programmer was male or female, young or old, black or white, rich or poor. No lofty title would make the code suddenly run (to my disappointment). Everybody had to do go through the same difficulties to make things run.

And while there were of course human biases and cultural heritage baked in many computer systems, the overall sense was that automated treatment was less prone to unfair partiality.

This historically impartial nature of computer system gave them a lot of credit. When we could "let the computer decide" there was less need for justification. It felt more objective. The outcome was easier to accept for everybody.

We are now faced with AI computer systems that are no longer deterministic, that are unable to track their decision path, and that are trained on data that contains huge amounts of biased, skewed, and opinionated prejudices. This is essentially impossible to avoid even with the best of intentions since even the very selection of training sources already constitutes a preference.

Imagine the cumulative effect when widely used AI bots provide answers that are just slightly skewed towards one commercial product, one news source, or one political topic. Imagine discrimination against certain groups of people. If so many individual people are slightly misled, the whole society shifts.

The bias can result from the training data, but it can even be deliberately put in the "harness" that provides access to the model. Many such attempts have already been reported, including political influence and undermining open community sources like Wikipedia.

False Attribution

From our history with computers we assign them the attribute of impartial, factual expertise. It is really hard to overcome this reflex. Incredibly hard. Let me give you an example.

At a recent high-profile developer conference, we discussed in a small group of leading experts. One asked "hey, what again does G-P-T exactly stand for in the word ChatGPT"? We had an expert in the group who knew the answer. He helped to develop that product. We knew he did. But only after someone asked ChatGPT the same question and got the same answer, everybody was satisfied.

Oh, the irony!

The logical fallacy of false attribution is technically a bit different from how I use it here. I know. Forgive me. It is somewhat related, though.

Think for Yourself

It is just so tempting to delegate decisions to a system that "talks" eloquently, has access to a huge amount of data, always sounds confident and uses a machine that in the past has been a reliable, impartial servant.

It takes insight to see through the false attribution. It takes courage to trust your own reasoning. It takes effort not to resign.

Even the best of actors with the best of intentions and the best expertise cannot make AI systems impartial and reliable. On top of that, there are bad actors that use conditioning to corrupt your mental faculties and then prey on the self-inflicted immaturity.

I know it's hard. But I hope that by recognizing our fallacy of false attribution within yourself and within others, we can keep the AI systems away from harmful outcomes like mass surveillance, mass political influence, mass economic dependence, and control over critical infrastructure.