Some Musings on Assistive AI
AI Generated Agent; Courtesy OSET Institute
This year, while I've been treading in the deep end of the AI pool, some observations are emerging about the realities of AI Agent deployment, principally because one of my clients is in the public service sector (the OSET Institute and it’s TrustTheVote® Project). The two other projects I’m engaged with are in the commercial sector; one as a global brand and the other a Sports-technology start-up venture. The trio provide an interesting spectrum of application of AI, from infinitely scalable call center agents, to a highly sophisticated athletic performance assistant, and the integration of AI concierges into the retail experience. The thru-line in all of them is a particular focus on the customer engagement experience as a component of a CRM strategy.
One thing becoming clear across that spectrum: we need to be suspicious of the claims of some “predictive” models. Actually, that could be an entire article posting of it own, so let me briefly say this about that: as best I can tell, at least for social and public sector applications, nearly everyone claiming to be able to apply predictive models to social behavior is being disingenuous: many of these algorithms do not, in fact, make predictions that are any better than guesswork. So, for at least organizations acting in the public best interest, as well as corporate brands considering the opportunity of AI in CRM/CES I recommend avoiding this area for now; so far, they are costly expeditions. Remarkable gains are being made with predictive modeling and machine learning, but not without some significant challenges. Again, this is all probably better saved for a different post.
However, I raise this here because in my work I am seeing some interest in AI Agents that can autonomously act on one’s behalf based on the Agent’s predictions of the individual’s wants or needs (and in the case of the public sector, acting on behalf of a citizen is downright dubious). This is the so-called automated AI, and at this point it requires predictive modeling where so far the juice is not worth the squeeze. While some brands in the corporate world may have the financial wherewithal to enjoy this intellectual frolic, I maintain that for the investment, AI can do a bunch for the customer engagement experience in an assistive capacity without needing to act autonomously on the customer’s behalf. In short, attempting to assume the behaviors or needs of one will conform to a model of many is a fool’s errand at this point.
So let me back up a bit (I did title this “musings” but I don’t intend just ramblings). An important element to leveraging AI for far higher CRM engagement will be differentiating assistive AI and automated AI.
Assistive AI helps human operators process and consume information, while leaving the human to take action on it.
Automated AI acts upon that information seldom with human oversight (e.g., the autonomous automobile).
The middle ground is the AI concierge the customer has authorized to act on her behalf and refill or replace product when the Agent determines she is likely in need.
For the near term, assistive AI should be advance and leveraged first, in order to gain early brand or organizational "design wins", while moving slower on automated AI.
Let Me Explain Some More Why
The challenge as I see it now is that AI algorithms are all too often encoding human bias.
And at least in the public sector (and in some corporate brand marketing situations), failure carries real life customer consequences. Of course, in the private sector, brands can decide that a certain failure rate is acceptable and let the algorithm perform. (That’s not the case for public sector facing organization.)
On the other hand, we must keep in mind that when customers interact with brands, similar in some ways to how citizens interact with their governments, they expect fairness. Since inherent "AI judgement" (the bias) may always be in play, automated AI cannot yet guarantee to deliver on brand promises. That’s OK; we’re still scratching the surface of conversational AI and assistive AI services.
There is plenty of headroom in advancing assistive AI via conversational agents. Indeed, with technologies like Soundstorm (from Google Research), and the emerging ability to produce conversational agents by training them on disfluencies, the human quality of value-based conversations can galvanize customer experience by delighting the interaction (which will tend to catalyze up-sell, cross-sell, and loyalty, if not advocacy.) So, whether customers are ready for (or expect) autonomous agents to serve their every whim is not a priority (although getting bias out of algorithms, in general, should be).
A Mind-Numbing Example of Advances in Conversational AI
I have an incredible example of this natural language advancement contained in this 14 minute audio link to a podcast explaining how all this works, with one wild twist: the individuals conversing on this podcast clip are not human beings, but actually AI agents who are interacting with each other… both having been "primed" with an entire blog post by computer scientist Simon Willison, which is used as a prompt to these two AI agents, which in turn, triggers and entirely AI generated conversation between them. Prepare to be blown off you chair listening to this.
So, how does any of that matter to integrating AI into CRM strategies? Because while automated AI looks slick (it’s not really), assistive AI with this level of conversational capability should be more than enough for brands to take some impressive, cost effective, infinitely scalable leaps forward in customer engagement.
The bottom line is there is plenty of headroom in assistive AI for CRM, and therefore, plenty of lead time to work out the bugs (like algorithmic bias) before the emerging requirement of automated AI darkens the C-suite’s doorstep.
Reach out to us to learn more; this is incredibly exciting next generation stuff that’s here now, and ready to move CRM to the next level of highly personalized customer/consumer engagement.