I stumbled across this: The real Threat of AI. Read it ICYMI, it’s a great op-ed.

Then again, I think it buys a common misconception that causes it to sort of miss the point. Here’s why.

The idea of AI (I use the label inclusively to mean AI, ML, DL and other alphabet soup, tech details are irrelevant here) as colonialist rent extraction rests on the assumption that a country’s training data are free for the taking.

This is utter, complete, blatant bullshit.

I can understand why important economic interests are pushing this bullshit, but bullshit it is.

Now, maybe we should stop considering AI as “technology”, something you go and buy. Modern ML-based AIs are basically generative code trained on massive amounts of data. The original code is only a potential AI. No data, no AI. You just dont go out and buy AI. You grow some potential AI on proper data. The more and better the data, the better the result.

This should make clear the misconception: a trained AI depends as much on the data as it does on the original code.

So, if some digital giant wants to train its AIs on a country’s data, the deal is simple:

  1. data is paid a fair price, unless they are open data in which case they may be free
  2. regardless of 1), ownership is split 50/50 for any resulting application
  3. appropriate taxes are levied on the operation of the application to the country’s citizen; no fiscal loopholes
  4. in the case of health data, it makes sense that all servers reside within the data’s country of origin; just to be on the safe side.

I am aware that this approach makes it hard to turn initial investments in AI into a tool for perpetual, colonial-style rent extraction. It would also be harder to turn those initial investments into billions of dollars for the original VC, as operational costs would be far higher.

Guess what, I could not care less. This way of dealing does not keep investors from making money, and is in line with the “data is the new oil” idea: it would generate permanent value also for those people and countries who create and own those data, not only for AI would-be colonialists. I am aware this is not how it’s worked so far with actual oil. Still, I could not care less.

As Taleb’s Fat Tony would have it: Gotta problem widdat?

Share This

Share This

Share this post with your friends!