Taking a Trip with Hallucinatory AI
I admit that I’m still struggling to get the hang of retirement. I’m officially retired – in that I no longer collect salaried remuneration from an employer for professional services rendered – but I’m not exactly living a superannuated life of leisure. No beach chairs, sun parasols, golf clubs, private-club memberships, retirement homes, and lawn bowling are included in my present existence, and I’d have to strain hard to envision them becoming fixtures in my future.
I’m always trying to learn something new, curious about and engaged with the world as I find it. For me, retirement does not mean stopping in any sense. It just means that I have more free time to spend doing what I want to do rather than what I must do to satisfy contractual obligations. As long as you stay active, and you remain involved in furthering your knowledge and understanding of the observable world, you’ll find that retirement expands your horizons rather than narrows them. You’re able to push into new areas on your own volition, evaluating them on your own terms.
The foregoing was prologue, much like the past.
Now let’s talk about large language models (LLMs) and generative AI. I’ve been experimenting with a few LLM-based chatbots, submitting some simple and clear prompts to see how well genAI can deliver on its off-the-hook hype. The short answer: The promise remains, but there’s a lot of work to be done before we’re at a point where we can depend on the technology to provide accurate and actionable information that we can trust implicitly. AI isn’t trying to lie, but nor is it under any compunction to always tell the truth.
I arrived at this conclusion directly after submitting an identical set of queries to different chatbots. I asked each to provide information on the hydrogen economy (a topic I’d been researching lately), companies active in hydrogen technologies, why public companies choose to leave the London Stock Exchange, and, finally, a brief biographical sketch of yours truly. I made the latter request not to gratify a narcissistic compulsion, but because I assumed, correctly as it turned out, that my obscurity might provide a challenge to any intelligence, artificial or otherwise.
The chatbots generally did well in describing the hydrogen economy, each providing similar definitions and explanations of what it represented, what it sought to address, what challenges it faced, and its current disposition. Despite quibbles here and there, I found the information received was reasonably accurate. The information was good enough, in fact, for me to posit tentatively that AI chatbots might be employed gainfully for basic discovery research, though there were enough minor ambiguities, omissions, and misstatements to require the continued service of post-prompt, human-mediated fact-checking. You cannot assume that chatbots are omniscient because, well, they’re not.
My next prompt, for information on companies active in hydrogen technologies, elicited responses that were mostly accurate, though some of the companies cited had greater involvement with hydrogen technologies than others. Again, this sort of information could provide the basis for further research, but you’d be remiss to trust it completely.
I also thought the chatbots performed well in addressing the question of why companies have chosen to leave the London Stock Exchange. The chatbots differed slightly in their prioritization of the causes, but they all cited a consistent set of relevant factors. Again, this sort of discovery research is a good application for genAI, but fact-checking remains essential.
More intriguing was what the AI chatbots had to say about yours truly. My prompt was spare and simple: “Tell me about Brad Casemore.”
Hallucinating a Parallel Existence
You might be discouraged to learn that there are a few Brad Casemores on the loose, but thankfully not many. I could tell from the chatbots’ answers that they were endeavoring to describe me rather than others who share my name. That said, the hallucination quotient was vertiginous, to such an extent that I began to wonder whether I had a Jekyll and Hyde alter ego living a parallel existence.
For one thing, the chatbots informed me that I was still gainfully employed – not retired at all – and working at companies where I have never been employed. Here’s what one chatbot offered on the subject:
Brad Casemore is a renowned technology executive and investor with a notable career in the tech industry. Currently, he serves as the President and Chief Executive Officer of сборка Systems Inc., a software company specializing in cloud-native application delivery. This role has made him a prominent figure in the enterprise software and cloud computing space.
Prior to his role at Сборка, Casemore held numerous senior positions at prominent tech companies. He was the Chief Operating Officer at VMware, a leading virtualization software company, where he played a pivotal role in its growth and global expansion. During his tenure, VMware experienced substantial growth and became a significant force in the enterprise software market.
Before joining VMware, Casemore held various leadership positions at Microsoft, including serving as the Corporate Vice President for the .NET platform and developer tools. He was instrumental in the development and launch of Microsoft's .NET framework, a key initiative in the company's strategy.
Casemore's extensive experience in the tech industry has also led him to serve as a board member for multiple companies, includingなり複雑ができる。 He has provided his expertise and guidance to emerging technology businesses, contributing to their growth and success.
His insights and contributions to the industry have been recognized, and Casemore has been featured on the Forbes Technology Council and named a “Tech Titan” by CRN Magazine.
With a distinguished career spanning multiple tech giants, Brad Casemore is a respected leader and a valuable asset to the technology industry. His work has been instrumental in shaping the modern tech landscape, and his contributions continue to influence the direction of enterprise software and cloud computing.
None of that is true – not for this Brad Casemore, nor for any other Brad Casemore of which I am cognizant. It’s confabulation, or hallucination, on a grand scale, but not without its entertainment value. Among the many amusing highlights, I would like to draw special attention to the Japanese characters, which apparently translate into a company name that reads, “It can be complicated.” Indeed, I agree, it can be complicated, even to the point of absurdity, not that I'm complaining.
Another spin of the chatbot wheel of fortune informed me that I’d written several books, including “The IT Services Marketing Handbook” and “IT Services: Growth, Scalability, and M&A Strategies.”
Don't Trust, Always Verify
Dear reader, I have not written those books, nor am I the “founder and principal analyst at BC Strategy,” a company with which I have no affiliation whatsoever but one that the chatbot somehow linked to me. I’m not Sherlock Holmes, but perhaps my initials caused the chatbot to make that mistaken connection.
As for the books, I would very much like to write one, but I’m thinking more along the lines of a satiric novel. Perhaps the chatbots will attribute such a literary work to me so that I don’t have to go through the weeks and months of sitting at a keyboard and, you know, actually writing. You can do it, chatbots! You’ve already got me writing books; all you need to do now is put me in a different genre.
One chatbot, though – which, like all the others mentioned here, shall remain nameless because I realize the technology is nascent and likely to improve – captured me in all my glorious imperfections and opacity:
As of my last update in January 2022, there isn't a widely recognized figure named Brad Casemore in any prominent fields such as entertainment, sports, politics, or academia. It's possible that he could be a private individual or a professional in a niche field not widely covered in public sources up to that point. If Brad Casemore has become notable or gained recognition after 2022, I wouldn't have information on him beyond that date. If he's a private individual, I wouldn't have any information about him, as I don't have access to personal data about individuals unless it has been shared with me in the course of our conversation.
This is in the same neighborhood as the truth. The chatbot, to its credit, doesn’t reach beyond its grasp. From my perspective, admittedly, it’s not a lot to leave to posterity, but I can’t blame the chatbot for that.
So, why do LLMs hallucinate like Ken Kesey his Merry Pranksters dropping LSD on the psychedelic bus they called Further? An article published last fall in Psychology Today offered the following explanation:
The short answer is that LLMs have no innate conception of truth or falsehood. They use a purely statistical model of human language created from news stories, blog posts, ebooks, and other human-created text online. This statistical model allows LLMs to guess the next word after a prompt and to do so repeatedly to generate an answer or an essay.
That’s helpful, to a point, but why do LLMs confabulate when short of hard data? Well, one answer is that while LLMs are not human, they are created and informed by human knowledge and indirectly subject to human frailties. The Psychology Today article concluded with a pithy observation: “Chatbots do not have egos, but they imitate humans who do.”
This might explain why they are prone to embellishment and rationalization when the informational trail reaches a sudden end. That is also why, as you and I begin using the technology for practical purposes, we should modify and update a hallowed maxim to govern our actions: “Do not trust, and definitely verify.”
Therein lies the rub. Until we have more trust in the accuracy, integrity, and quality of AI’s bounty, the technology will not represent the all-knowing, time-saving “easy button” that many want it to be.