General idea
With your daily experience with ChatGPT, Goolge, and other AI tools we know to this day, what are your thoughts? Are they incredible? Are they better than you are? Are you amazed by it? If you like me, I think AI is incredible, it is much better than I am.
The focus of this topic is on AI safety. The approach is the common sense - not the claim of any famous person, not the word from the people you admired, not a word from an authority or words from credit researcher. At this point in time, yes, there are many companies that doing the frontier research and all claimed that they are play safe in sandbox settings. On the other hand, we know there have been applications that allow an AI centered program to actually control important facilities that ranging from company's website, company's computers, company's network, which, can interact with world wide network.
It is true that words from frontier research comany is that the AI haven't reach the AGI (artificial general intelligence). In other words, it is not quite reach what a human can do yet. But we also know, there been cases report that the AI were not truthful with the researcher, the controller. What this mean is that even the best researcher could not control the whole aspect of the AI.
Now, let's look back of our own experiences. Have you expericed or heard of that ChatGPT or comapy Chatbot replied with sily answers? I think we all have. Most of us don't think much about it and just laugh those off. But, think about this: These are PRODUCTS that companies put out and used in daily life. So, the important point is this: these are IM-PERFECT but ARE exist and used in the real life. Imaging this: what if this is not a chatbot, but an AI program that handling your back account? Do you trust the IM-PERFECT product?
The point now narrowed to why we can just laughed off these instances? Well, the fundamental reason is that these are not a large or life-threatening damages. For example, even if your bank accout been stealed of millions of dollars, the damage is on you and the bank. It is considered controllable and not affecting a majority of people. But I have to emphasize that the limited damaging is really not the point here. The point here is that people can't well CONTROL the AI yet. But people already using it in the real life and there is no rule or regulations that limited the use. Without the rule or regulartion in place means anyone can build and put it into use - for good or for bad. Yes, we do have other law that can control the damage - like the court will ordered the bank to return your money. But this is only beacuse the AI hasn't damage other part of the society. What if that AI application also destroyed the court system?
So the point here really is the uncertainty of the AI application - people don't know how to control it precisely and, also, it has not have enough permissions to access important facilities/resouces. But ! What if people, without knowing how to control the AI precisely, but allows the AI to access more resources? What to prevent people of doing this? From people's experiences with ChtGPT and other AI tools, we ALL believe that AI is incredible and smart than us. It is hard for us not to allow it to access all the resource we have access to given that we want it to help us do what we want to do. So, bank WILL allow it to access accounts.
So, does the risk exist? I think we can all agree that yes, it exist even if the AI has not reach the AGI. It really don't matter if the AI has reach the AGI, the risk is already there and, we haven't been able to well control the AI of today, let alone, the AI of the future.
WE OF THESE DAYS, should already thinking about rule and regulations - how to limited the use of AI for companies - so the damage can be controlled - like no access to external networks.