Another Bad Critique of AI, in the New York Times

Want to get your uninformed opinion about AI published in the most prestigious newspaper in the United States? Then follow my advice.

First, it helps if you have little or no perceptible training that would give you the necessary background to critique AI. For example, perhaps you majored in history, instead of relevant fields like computer science, or philosophy, or neuroscience. Or maybe you went on to study something even more unrelated; let's say, getting a master's degree in business administration. Why should you know anything about the subject you're critiquing? The New York Times certainly won't care, and neither will your readers.

Second, have a vague thesis, perhaps something like humans have capabilities machines will never have. Be sure to support that thesis with unbacked assertions, confidently proclaiming things like "judgment cannot be automated" and "Judgment ... is a uniquely human skill". It is important to not specify how you know claims like this with such certainty. Do not refer to any published studies or journal articles backing up your claims. Uniquely human? Don't refer to the vast ethological literature, where examples of judgment in non-human animals have been documented again and again. And above all, do not point to all the prior claims about abilities machines will never, ever have---claims that have been proved wrong over and over again!

Third, you should relate some anecdotes. It's important to have anecdotes, rather than any deep investigation or hard data, because people love stories, especially stories that reinforce their prejudices. But it's also important to have those anecdotes as hazy as possible, so that no one can check the details. For example, just talk vaguely about the failures of an "A.I.-assisted analysis" or "an A.I. model" without specifying precisely what software was used and when. After all, there hasn't been very much progress in AI lately. Two anecdotes should be more than enough.

Be sure not to explore any alternative explanations of why the unnamed "A.I. model" failed to give useful advice in your particular situation--that would be too scientific. For example, if it was because "The chief executives of the two companies had a fraught relationship", be sure not to specify if this crucial fact was included in the training data. This will allow you to triumphantly conclude that "The model could not account for the interpersonal dynamics".

Finally, if your argument consists mostly of things like (in my words) `here are a couple of times where AI gave the wrong answer to my business issue', certainly do not mention that humans give wrong answers to business issues all the time. It is vital not to refer to any data-based comparison of the capabilities of humans and machines, because that might cast doubt on your thesis.

If you follow my advice, millions of people will read your timeless prose. And afterwards, you can get back to the vital work of investment banking.