Where I stand on AI
I’m a member of BNI Excel, and our chapter recently had a lively discussion on Generative AI for marketing purposes, primarily on social media. Without recapping the whole thing, we have a mix of people who are super-bullish on AI and people who are skeptical. As for me, which matters because it will extend to how this Firm interacts with you? I’m interested in the risk side of this conversation.
Let’s start with the basics. These are large language models that have been trained to resemble a person - and we’ve had these in one form or the other for quite some time. There are important technical differences between things like IBM’s Watson, Siri/Alexa, the predictive models and automation that were all the rage in the 2010s, and generative AI, but there is no magic here. Computers learned how to play chess with predictive algorithms that were trained on hundreds of thousands of chess games. The same concept is what drives modern Large Language Models (LLMs) to emulate us, just in a far less structured arena.
Because of this, there are a few points of AI risk that often get lost in excited conversations about what it can do.
Data Privacy. Because LLMs have a voracious hunger for data, they may be collecting the information that you give them and reusing it without your consent. This is why a lot of banks and financial institutions use CoPilot instead of something using more powerful models. CoPilot doesn’t scrape organizational data according to its privacy policy. It meanwhile will not let you use Anthropic’s models in the EU because it cannot assure compliance with the General Data Protection Regulation (GDPR), and for that matter its implementation there may be causing a lot of companies to fall afoul of the regulation, to the tune of up to 2% of a company’s global revenues. Before using a model with information that might be potentially sensitive, especially if it includes Personally Identifiable Information (PII), you should carefully review the privacy and use policies of whoever is hosting your model.
Hallucination and Misrepresentation. Because the model is trying to have a conversation with you, it often makes things up. When I wrote the business plan for this firm for instance, I repeatedly caught the model “inferring” projections based on what turned out to be nothing when I interrogated the assumptions. More famous instances such as lawyers citing fake case law have led to professions, including mine, having to adopt rules reminding professionals of their due diligence requirements. Even if you are not held to such strict standards, anything that a model generates is your responsibility, and a potential avenue for loss.
Cognitive Debt. If you’ve done any coding, you might be familiar with technical debt. AI has given birth to its younger brother, cognitive debt, which is what happens when the AI is doing the work for you, but you don’t know how to reconstruct it yourself or even how it works. This can be as simple as not knowing something that a marketing agent promised a customer, to not understanding how to scale up or add to a functioning system that was built by and is used by agents. Debt of course is not a categorical bad thing, debt can help you grow for instance, but not managing your debt can be fatal to your enterprise. It’s no different with technical and cognitive debt.
None of this is a) exhaustive, b) a reason to swear off Artificial Intelligence, or c) material to say that Artificial Intelligence will not dramatically and durably change the way we work. It is to say that like all new technologies, AI carries risk and like all business risks, it needs to be assessed and met with controls sufficient to reduce that risk to an acceptable level.
By the way, if you need help with this assessment, and with designing controls that can meet this and other risks impacting your business. Let’s talk.
