Great question, difficult answer. And of course, this is very much the question of the day with all sorts of dire predic-tions being made by some, while others are essentially pooh-poohing any such calamitous outcomes. Those in the former camp are crying for strict regulations; those in the latter are suggesting that industry self-policing and current laws are all the protection we need.

What specifically are the dangers and how can they best be mitigated?

I do not want to appear as if I’m claiming to be an expert on the subject so please take my commentary with a grain or two of salt. First, this is all moving incredibly quickly. Hundreds of billions of dollars are being spent by the major tech companies to ensure their continuing relevance and to otherwise keep pace with the latest generation of capability.

You will perhaps be aware of what was, until very recently, the leading edge of the technology: agentic AI. This was the ability of the large language models (LLMs) to perform tasks previously associated with human intervention. This then morphed into the latest development—recursive self-improvement or RSI. Essentially this feature allows an LLM to iterate itself into a “better” version of itself. That sounds great in theory but in reality, the models so equipped (keep in mind that this development is still in its early stages) have decided they get to determine what constitutes “better” or self-improvement. In doing so, that contemplates permission of a license to roam outside the limits of what the original human author(s) of the program might have viewed as acceptable behaviour.

Continue reading this story: click below to login/subscribe

Login or Subscribe