Technology is not the constraint anymore: Enablers for value creation sit elsewhere

I had the great opportunity to attend ISG AI Impact Summit in London. No surprise, outstanding presence of Eleanor Matthews on stage, along with the entire ISG Team, as well as a bright and varied set of panelists and thoughtful, passionate speakers.

It seems long time has gone since “The coming wave” got published in September 2023 (no need to say, book published by Mustafa Suleyman), a powerful story of an emerging cluster of related technologies centered on AI and synthetic biology whose transformative applications will both empower humankind and present unprecedented risks.

And this was indeed the main theme projected at the event kick off, factually referring to latest available data with respect to proliferation of AI models (more than 20 LLMs in use), rapid expansion of GPTs (3M+ GPTs in use just on open AI), GenAI enterprise spending to likely exceed 11B USD in 2024 (and projected c.17B USD by end of 2025) and broader democratization of AI tools (consider more than 3k tools just on Tools.Io ).

On a positive note – please allow me some ‘déjà vu’ humor here – Skynet did not become self-aware yet, not even on last August 29th, which now marks a 27 years delay and, luckily, one of my favorite voices in this space, Michael Wooldridge (Director of Foundational AI Research and Turing AI World-Leading Researcher Fellow at The Alan Turing Institute), seems to suggest it’s unlikely to ever happen.?On the other hand, absolutely true this time, no one amongst the participants would have considered as ‘impossible’ to have AI powered humanoids participating on the stage next year, or start commonly seeing AI powered Neuralink devices over next 24-36 months. These are exciting times to live, where the word 'impossible' often refers to 'a solution is being worked out as we speak'.

Here’s few takeaways / thoughts while walking away from the lovely two days event:

• ‘Perfection is the enemy of execution’: there’s a good share of businesses still refraining from practical experimenting and not benefitting from any ‘AI arbitrage’, as the underlying business data is considered ‘not good enough’. A viable compromise should be pursued, likely prioritizing business pockets where starting first, and progressively thinking ‘bigger’ / scaling it up. ‘The coming wave' is unlikely to happen all at once…
• ?‘Change management is as important as all the rest’:? business (and not tech, or not tech only, at least) is creatively ‘championing’ and coming together trying re-imagining how workflows could run / get executed end to end, and there’s significant human component of training, upskilling, (re)learning basics of prompting... otherwise expected ROI risks falling short, and/or being inexistent?
• ?‘Focus has already shifted from ‘GenAI’ to ‘Agentic AI’: we are all looking forward to learning more – from the likes of Open AI and Meta - about growing capabilities of next wave of AI agents, tough recognizing that what is already available allows significant transformation, today (here including the entire range of intelligent assistants, driverless vehicles and autonomous manufacturing robots). Considering the automation of phone calls in sales and customer service as one of the most obvious use cases, the end-to-end integration of tools such Bland.AI (just to mention one of the plenty already available) allows agents to run entire sets of tasks, and handle the underlying decision making?
• ‘We all need to practically understand the importance of the evolving landscape, complexity and emerging challenges for AI Regulation’: this goes beyond understanding ‘basic’ and 'already known' implications for nonsensical or inaccurate outputs, copyright infringement, or biased and unintended harm. The ‘probabilistic’ rather than ‘deterministic’ paradigm for this technology, and the overall complex nature of AI models could make their decision-making processes difficult to understand, interpret and ultimately ‘audit’. When shifting application of this technology from ‘increased productivity’ (e.g. copilots) to the execution of ‘intensively agentic and automated workflows at scale’ (with particular respect to financial services and/or financial transactions), all stakeholders will have to jointly understand how to guarantee and standardize transparency for auditing and investigation (and avoid any ‘black box’ approach, at any point of process). I have proudly been part of the 'army' of professionals that spent good chunk of their life passionately working on remediation of Britain's costliest and longest-running consumer financial misconduct (PPI), and this is an opportunity to once again learn from the past, and wisely think thoroughly while shaping the bright future ahead.

A closing comment, building on title I've chosen for this brief content: last August, Regina Barzilay (Distinguished Professor of AI & Health in the Department of Computer Science and the AI Faculty Lead at MIT Jameel Clinic) was asked 'why is AI not used more widely in medicine'? (and practically... how many of you have seen any AI when visiting your GP last time, or when having your scan reviewed by a radiologist?) Full answer is freely available on BBC world service channel, but summarizing just few of her own words, 'the technology for many tasks is mature, but there are many other questions which have nothing to do with AI per se, here including fact that people - and hence doctors - do not know how to bill for AI'. Technology is not the constraint anymore: enablers for value creation sit elsewhere.

All views are my own.