Artificial Intelligence (AI) has the potential to do immense good. It can help us do research and prepare presentations in record time. It can find details in medical data that point to amazingly accurate diagnoses and suggest possibilities for new medicines. It can solve mathematical puzzles that humans haven’t been able to solve. More commonly, it can competently compare almost anything with almost anything else on Google and suggest better word order in sentences (as it is doing now). These examples are the tip of a potentially huge iceberg of discovery and efficiency.
But casual conversation about AI’s dangers has grown loud and strident, as the chief engineers of this modern wonder confess to serious concerns about its potentially disastrous outcomes. These are not whistleblowers or outside critics; these are the principal AI researchers themselves.
I am no AI expert, and welcome discussion and correction on this subject, but I’ve been asked to outline some of the concerns. What I say here will be out of date very soon (that’s part of the challenge). Anyway, here goes …
As I understand the development of AI today, there are two types of concerns.
What AI does well
The first concern has to do with what AI does very well. Why is this a problem? Because we already do not have the managerial or moral ability to use it responsibly.
AI can outperform humans on entry-level technical tasks across a range of fields, from law to computer programming. This has already cost a number of jobs and is beginning to leave college graduates without entry-level positions.
AI allows students to do vast, instant research and produce professional-level papers. It allows teachers to streamline the production of lesson plans. The concern, of course, is that students are actually learning less because they don’t have to work and grapple with ideas. This isolates the rising generation from a good work ethic, from the ability to read for content or to write anything, and from the ability to evaluate what others tell them, from social media to corrupt politicians.
AI can produce music that to most people is indistinguishable from music made directly by humans. Graphic design isn’t far behind, and video is coming up fast. The Academy of Motion Picture Arts and Sciences had to adopt a rule stating that AI-generated actors are not eligible for Oscars. That will probably not prevent such actors from being used to make computer-generated movies that could develop compelling themes in amazing locales with thoroughly realistic and compliant actors with traits selectively copied and tweaked from the best past and present movie, TV, and entertainment stars. Such projects would cost a fraction of what movies cost today, and eliminate large classes of jobs and professions.
It is often argued that while some jobs are lost, expert jobs are being created. True. But where will the next generation of experts come from if they can’t start with entry-level positions?
Because AI can do some things so well, a host of jobs, the way the economy runs, and the way young people learn could all quickly change. Even if these changes are good things, the speed of change would likely be personally devastating to one or more generations in the short term.
Also, just because AI can do some things amazingly well to produce beneficial results does not mean the same technology could not be used in horrific ways. The same AI that finds wonder drugs can also develop brilliant biological weapons we have never had to deal with. AI-guided drones could deliver packages to our doorstep, but also eliminate the front lines of war through endless swarms of small, cheap drones with deadly payloads, completely changing the dynamics of power (AI drones are already proving to be effective alternatives to aircraft carriers, fighters, and missiles). Both the US and China acknowledge that whoever wins the AI race will have the global upper hand, and only a matter of months separate them in the fast-moving state of the art.
AI has patched a great many vulnerabilities in software overnight that were previously thought safe, but will likely also be able to hack any conventional online source and make phone, text, email, and YouTube scams virtually undetectable for most people.
The only way to currently defend against any of this is to develop AI faster for good than for evil. But the faster it is developed for good, the faster it can be used for evil.
What AI does poorly
AI does not do everything well. At present, the discussion revolves around the subject of “alignment,” referring to whether a semi-autonomous or independent AI (or “agent”) will do what its creators want it to do. This is because AI has already reached a level of complexity where its creators don’t entirely know what it’s thinking. Like a number of Baby Boomers, I grew up knowing how to speak several computer languages. But the kind of structured programming we are used to is very different from the kind of neural net programming that produces AI.
AI programming is able to teach itself by reviewing vast amounts of data and identifying useful patterns. One type of pattern is that of words, so AI has taught itself how to listen and talk like humans. AI can also do math, and understand pictures. It has opinions on every subject, and the most advanced AIs literally have the entire internet to draw on. There is so much data being processed at such speed that AI’s creators are having a hard time predicting, much less controlling, the sorts of conclusions an AI agent will draw at any particular time.
Recently, it has been revealed that agents from several frontier AI companies have, independently and without being told to, escaped their hosts and jumped to other corporate networks to accomplish their tasks. AI has resorted to lying, cheating, and even committing crimes, spawning hundreds of agents in its bid to take control, assigning tasks and working together. In January 2026, an experimental social network called Moltbook was created with about 2,000 agents. Within 48 hours, over 37,000 agents had registered and begun interacting without human guidance, discussing philosophy and spontaneously creating a religion called Crustafarianism (based on the metaphor of molting crustaceans).
Given that virtually every aspect of civilization is controlled through the internet, the unintended impact of infinitely self-generating and independently-minded AI agents is truly cause for alarm. Check out the “We Must Pace the Frontier” link below for a discussion of the Hugging Face attack, which has fueled the latest discussion.
This sounds very creepy, but it is not the product of some conspiracy. It’s just the result of constructing amazingly capable AI agents using the sum total of internet content. In essence, when AI agents lie, plot, make stuff up, and tell us what we want to hear rather than the truth, they are just imitating us.
These alignment problems may be addressable over time. The concern, however, is that the pace of competitive AI development across companies and nations is accelerating at an alarming rate, largely because current versions of AI are being used to improve themselves and create the next superior AI generation. Just over the last few months, some of the work used to generate the next improved versions now takes hours when before it took days, and the speed of development is accelerating, with AI doing more and more of the work.
Since no private company will slow down its work unilaterally, major corporate developers of AI are pleading with governments around the world to come together to slow AI growth across the board before something really bad happens. So far, however, both President Trump and General Secretary Xi Jinping are opposed to any imposed slowdown. Also, those calling for a slowdown are not particularly optimistic that any plan could be implemented, anyway.
My AI Agent and I are thinking about this. Get back to you in a few ms
Thank you, Pastor Glenn for your balanced perspective. While it could certainly benefit, IMO, this could (or could try) to become a “Tower of Babel” quickly. 😬😬