Are there any genuine benefits to AI?

Oh I am in fact giving the giant auto complete function little credit. But just like any computer system, an AI can reflect the biases of it's creators and dataset. Similarly, the computer can only give an answer to the question it has been asked.

Dataset wise, we don't know exactly what the bot was trained on, other than "a lot". I would like to hope it's creators acted in good judgement, but as creators/maintainers of the AI, there may be an inherent (even if unintentional) bias towards the creation and adoption of AI. Just like how some speech recognition models have issues with some dialects or image recognition has issues with some skin tones - both based on the datasets they ingested.

The question itself invites at least some bias and only asks for benefits. I work in IT, and I see this situation all the time with the questions some people have in tickets: the question will be "how do I do x", and while x is a perfectly reasonable thing for someone to want to do, it's not really the final answer. As reasoning humans, we can also take the context of a question to provide additional details without blindly reciting information from the first few lmgtfy results.

(Stop reading here if you don't want a ramble)

---

AI is growing yes and it's getting better, but it's still a very immature field. Many of its beneficial cases have serious drawbacks that mean it should NOT be "given full control of a starship", so to speak.

• Driverless cars still need very good markings on the road to stay in lane, but a human has better pattern matching to find lanes - even in a snow drift.
• Research queries are especially affected, with chatbots hallucinating references that don't exist despite being formatted correctly. To that specifically:
  • Two lawyers have been caught separately using chatbots for research and submitting their work without validating the answer. They were caught because they cited a case which supported their arguments but did not exist.
  • A chatbot trained to operate as a customer support representative invented a refund policy that did not exist. As decided by small claims court, the airline was forced to honor this policy
  • In an online forum while trying to determine if a piece of software had a specific functionality, I encountered a user who had copied the question into chatgpt and pasted the response. It was a command option that was exactly what I and the forum poster needed, but sadly did not exist. On further research, there was a bug report open for a few years to add this functionality that was not yet implemented
  • A coworker asked an LLM if a specific Windows powershell commands existed. It responded with documentation about a very nicely formatted command that was exactly what we needed, but alas did not exist. It had to be told that it was wrong four times before it gave us an answer that worked.

While OP's question is about the benefits, I think it's also important to talk about the drawbacks at the same time. All that information could be inadvertently filtered out. Would you blindly trust the health of you child or significant other to a chatbot that may or may not be hallucinating? Would you want your boss to fire you because the computer determined your recorded task time to resolution was low? What about all those dozens of people you helped in side chats that don't have tickets?

There's a great saying about not letting progress get in the way of perfection, meaning that we shouldn't get too caught on getting the last 10-20% of completion. But with decision making that can affect peoples' lives and livelihoods, we need to be damn sure the computer is going to make the right decision every time or not trust it to have full controls at all.

As the future currently stands, we still need humans constantly auditing the decisions of our computers (both standard procedural and AI) for safely's sake. All of those examples above could have been solved by a trained human gating the result. In the powershell case, my coworker was that person. If we're trusting the computers with at much decision making as that Bing answer proposes, the AI models need to be MUCH better trained at how to do their jobs than they currently are. Am I saying we should stop using and researching AI? No, but not enough people currently understand that these tools have incredibly rough edges and the ability for a human to verify answers is absolutely critical.

Lastly, are humans biased? Yes absolutely. You can probably see my own bias in the construction of this answer.