Failing to foresee the current state of AI
The last 14 months or so have seen incredible change in AI technology. AI has progressed beyond a level that many analysts thought it would take many years – or, indeed, many decades – to achieve. In this blog post, Scott Aaronson, who is a computer scientist at MIT, reflects on why he, too, failed to foresee how quickly AI technology would change. Essentially:
- His reasoning failed to take into consideration the possibility that neural network technology, which enables the recent developments that we are seeing, would take off and improve in the way that it did;
- Computer scientists don’t really “understand why deep learning works, in any way that would let us predict which capabilities will emerge at which scale”. If you don’t understand how something does what it does, it becomes incredibly difficult to predict where it head next and how it will get there.
It takes a lot of intellectual humility to recognise the flaws in our reasoning. What happens if those that lead OpenAI and other such companies lack this humility themselves, or are not surrounded by people that force them to engage in this kind of exercise?
The impact of AI on UK jobs
I have finally caught up with the “The impact of AI on UK jobs and training” report, published by the Department for Education. One interesting point in this report is that it looks at the impact of AI in general, as well as the impact of Large Language Models (LLMs) in particular. Though, it finds that the two are highly correlated. Here are the top 20 for each:
The other interesting point in this report is that it refers to research by the International Labor Organisation (ILO) which analyses the impact on jobs through AI-enabled automation (i.e., AI replaces the human) vs AI-enabled augmentation (i.e., there is still a role for the human, albeit it will be modified). The International Labor Organisation (ILO) predicts that the following jobs will be replaced by AI:
| Human resources administrative occupations |
| Book-keepers, payroll managers and wages clerks |
| Brokers |
| Market research interviewers |
| Pensions and insurance clerks and assistants |
| Authors, writers and translators |
| Telephone salespersons |
| Finance officers |
| Call and contact centre occupations |
| Other administrative occupations n.e.c. |
The ILO also predicts that the following jobs will be augmented by AI:
| Purchasing managers and directors |
| Education advisers and school inspectors |
| IT project and programme managers |
| Further education teaching professionals |
| Information technology and telecommunications professionals n.e.c. |
| Estate agents and auctioneers |
| Market and street traders and assistants |
| Physical scientists |
| Managers and directors in storage and warehousing |
| IT specialist managers |
Regulation of AI in the EU
Last week also saw the news that EU officials had reached a provisional deal on rules for AI technology. In the context of this rule, AI is defined as software that can “for a given set of human-defined objectives, generate outputs such as content, predictions, recommendations or decisions influencing the environments they interact with”.
The proposed legislation proposes 4 levels of risk for AI, with higher levels (e.g., facial recognition) seeing more strict controls over their use; and it attempts to legislate generative AI’s use of copyrighted material to train models. I found this short DW news piece a useful summary of what was agreed, as well as what remains to be done:
What interesting AI-news caught your attention, recently?
Ana Canhoto 