An article in ZDNET, written by David Gewirtz, says that we can harness the best of generative AI despite job disruption fears. “Generative AI has the potential to be a force multiplier for individuals working on projects. That said, it's hard to foresee whether AI will increase its value and threat, or find its value reduced – thereby also reducing its potential for world domination.”

This is undoubtedly a rapidly evolving field, Gewirtz says. “We've only dealt with generative AI and ChatGPT for just over a year, and we've seen the scope of improvement, pushback and ongoing problems.”

Gewirtz outlines some factors that could diminish the usefulness of these tools as we move forward:

1. Guardrails:As the owners of AI implementations discover that their creations respond in troubling ways, they're adding guardrails that limit the topics the AIs are willing to engage in. “This does prevent the AI from saying something heinous, but it also may limit the value of the AI as a content production tool.”
2. Limited response length:Currently, most AIs start breaking down if the response requires more than approximately 500 words.
3. Laziness:I have repeatedly tried to get various AIs to do aggregated research, examining as many sources on a given topic as possible, and constructing a literature review. But the AIs tend to look into their existing knowledge bases – and, scarily enough, Wikipedia – for the knowledge fueling their responses. “Today, there's no way to give an AI a research assignment and send it off to read all the literature on the topic. And, as such, there's no way to get an analysis of the results.”
4. Search engines optimizing for what is perceived to be human-generated content:That filtering is also true in search engines.
5. Data privacy:A big concern about public AI platforms is that the information passed into the AI becomes part of its overall knowledge base. “Anything considered confidential may not be provided to the AI; therefore, projects that require access to such confidential information may not have AI assistance.”
6. Intellectual property issues:This goes hand-in-hand with data privacy, in that the AI may produce content that creates liability for its users. “In my Halloween image generation test, for example, we noticed how much copyrighted content Dall-E 3 used to produce its output.”
7. Bias and fairness:While AI can help even the playing field for some who are challenged with writing, it also has a strong record of bias. “While there are many efforts to counter that bias, it's still a substantial problem – and is likely to remain a problem – because humans are inherently biased.”
8. Accuracy:As we've seen over the past year, AIs are capable of generating whopping big lies. “While there are techniques for reducing AI hallucinations, the overall concern about accuracy is quite valid.”
9. Adaption to breaking news situations:Sometimes our world changes instantly. “Whether it's a disaster we're suddenly facing or a business change that happened overnight, AIs may not have the necessary knowledge to respond accurately and relevantly. There's also a counterargument that when AIs are constantly sifting through live data, they may be able to notice changes and trigger alerts. This is particularly relevant in blocking cyberattacks.”
10. Regulatory and ethical governance:Depending on how worrisome AI gets, “we'll likely see regulatory or statutory defenses against AI behaviors, investment and use. It's too soon to tell how far that will go, but don't expect the AI freedom era of 2023 and 2024 to remain as wide open as we've become accustomed to.”

As with nearly all the technology previously created, AI has a dual nature, Gewirtz concludes. “It presents both challenges and opportunities. It's up to all of us where it goes. Will we be able to integrate AI into our lives as a force multiplier, or will we find ourselves fighting Skynet? Only time will tell.”

For much more, see AI taking on more work doesn't mean it replaces you. Here are 12 reasons to worry less | ZDNET.