By Anjan Roy
Amidst a fierce controversy over whether artificial intelligence companies should slow down and ponder about the larger implications of their work now raging in the cradle of their creativity, the Silicon Valley, some of the industry heavyweights are turning around.
Many in the IT and AI spheres are placing an alternative narrative and saying that there are enough checks and balances available already to rein in the runaway models creating wider problems. The biggest of them is the fear of liability from wider damages that any AI model can cause.
The corporations creating these high-power models and tools should consider the implications before releasing their products in the market.
Sam Altman of Open AI, who was one of the first whistleblowers to warn about the deleterious larger implications of the AI models, is saying that major corporations and players have built-in checks and balances to intervene and stop nefarious activities of “AI agents” well in advance.
Slowly, it is now becoming clearer as to what might have initially sparked the fervent talks about the impact of the independent AI agents playing havoc with broader frameworks.
One of the AI companies had created agents —units in the models which could work independently. These agents later were found to have discovered and identified some aspects of the functioning of another AI model and started attacking and destroying it. In doing so, the agents at a later stage even created swarms of other agents and all of them attacked this new model and brought it down completely.
It was then felt that if AI systems could be built, which could be so capable of acting independently of human command to identify and then create further agents and on their own destroy another model, then these could interfere with functioning of AI or computer networks elsewhere on their own. This is like when in an organism, a detective genestarts damaging the whole body and eventually destroys it.
For example, an AI model and its agents could start attacking a hospital computer network, or an airline operation systems, and bring down such facilities altogether. They can, in a similar fashion, go viral and keep on multiplying their destructive agents, which can then spread quickly into the entire interlinked computer systems.
The fear was about uncontrolled runaway spread of vicious malware wrecking widespread and uncontrolled damages. Say, it is possible to imagine a malware AI system entering into the defence and strategic systems of a country and setting off an uncontrollable chain reaction, releasing destructive power.
The sudden eruption of the controversy has in a way shaken the entire AI community and the corporations at the helm of these developments. The creators of these models themselves got scared of the Frankenstein type of developments of AI and their uncontrollable spread throughout the systems.
The corporations themselves are now realising that the scale of what can happen could in the process hurt the very originators themselves. Hence, they are talking of guardrails around AI systems for controlling and spreading such vicious aspects of their creations and work.
There has to be some checks and balances between the speed of development of AI models and products and the safeguards needed to be built around them.
Altman is now saying that while there are genuine grounds to be fearful about AI implications, there are some built-in brakes as well which could trigger corrective actions.
Yet, other industry leaders are talking of industry-level norms for regulations, rather than government-led regulations. The controversy is simmering.(IPA Service)
