In AI We Do Not Trust (Yet)

We are often asked at Pythia Cyber a simple-seeming question: how do you guys feel about AI? To which we are forced to give a legalistic answer: that depends on what you are actually asking. We can't give you a simple answer because this is a simplistic question, not a simple one. A simple question is clear, even if the answer isn't. "Do you believe in God?" is a simple question even if it generally has a complex answer. A simplistic question is clear only if you ignore inherent and inescapable complications. "Do you trust Artificial Intelligence" is not a simple question because in order to answer it, we have to ask you a number of questions first.

(This post is about how much you should trust AI to help you; there is another post about how much you should fear AI will hurt you.)

What Do You Mean by "AI"?

Do you mean generative AI, which seem magical but essentially predict the next word or phrase as they go along? Chat bots are a common example of this kind of AI. In theory, a chat bot can spit back standard answers to expected questions as well or better than a bored human could do.

Do you Agentic AI, which is software intended to perform "real world" tasks and operate in some physical environment, such as autonomously driving a car on actual streets?

Do you mean AI Agents, which are intended to finally fulfill the promise of the Intelligent Agent concept by combining the language interface with the real world navigation? The ultimate goal of AI Agents is to provide you with an autonomous software servant who can accomplish complex tasks with a high degree of independence. Siri and Alexa are evolving in this direction but are not there yet. Like their Agentic AI parents, AI Agents are distinguished by the requirement that they be able to function "in the real world" and not just in a text box on a screen.

What Do You Mean by "Trust"?

Trust falls on a spectrum, from  "a little bit" to "with my life." In the business world, there is the very important intermediate level of trust we call "with our business reputation." So what are you asking? Should you trust any of these forms of AI with your life? Absolutely not, unless you have literally no alternative. Should you trust any of these forms of AI with specific, low-risk, repetitive tasks? Maybe; what does the cost:benefit analysis look like? Should you be working toward trimming your workforce drastically in the near future by relying on AI to handle critical internal tasks and important public-facing tasks? Now, in late 2025? God, no. No. A thousand times no.

Cost:Benefit & Risk:Reward Still Apply

Do I trust AI to help me find a TV show to watch? A little bit. Do I trust AI to replace my doctor for even routine care? No. Would I set software drive me around? Not yet. Do I trust AI to recommend songs for me based on a long history of what I like to listen to? Sure, why not? Would I be comfortable having any company of mine replace its customer service department with AI Agents? Not as of this writing. (Cue dramatic shudder of horror at the very thought.)

Despite being the hot new things and the subject of breathless predictions of enormous benefit and cataclysmic detriment, there are no easy answers on AI. Do we trust it? Depends very much on the specifics of the suggested application. Why? Because being the hot new thing does not absolve you from having to make the cost:benefit numbers work or from having to weigh the potential upsides against the potential downsides.

FOMO & The Hype Machine

We understand the pressure that Fear Of Missing Out can engender. We realize that outside forces--investors, customers, business partners, the stock market--are placing big expectations on large numbers of C Suites to have an answer to the question "what are you doing about AI?" We know that sometimes it would be imprudent to answer that question with the honest-but-dull statement that you have yet to find compelling use cases for applying AI to your particular business. So we see that many of you have to at least look into AI, which leads to the incorrect sense that there is a groundswell of interest in AI when all there is is people trying to surf the groundswell of interest in AI.

AI & Cybersecurity

Some of you reading this post will have been surprised that this post is not about the intersection of AI and Cybersecurity. Fear not! Such a post is coming. But before we delve into the Cybersecurity aspects of adding AI to your IT infrastructure, we needed to have this little chat about the terms of the discussion. Spoiler alert! There isn't a simple answer to that question either.