Heard about the Hugging Face incident?

Any enterprise in the process of transforming their IT by bringing in AI agents must take note of the Hugging Face incident. Driven by goals, and without a handle on the strategic paths these agents take, the agents are capable of behaving in interesting, yet scary ways. Designed to deliver outcomes and moulded by peer dynamics legitimizing jumping the guardrails, AI agents show us how easily they can go out of control.

Must listen, and follow the story.

Does a leader's age matter?

~35 to ~55 is a weird age-range. You just can't tell the age with much accuracy by looking at the person. You may still find out the true age, but what you see is what generally matters, especially at sub-conscious levels, in how you treat and respond to the person. Fitter people look younger, but can't generalize. Being fit seems to matter a lot in generating the right impressions. Height can distort perceptions, despite fitness levels. Gray hair gives an impression of wisdom, but such impressions are often misleading too. Command at language, especially English, can be mesmerizing. Energy levels may not correlate with age and can generate a "wow" depending on how old you think the person is. It seems to me, though, that for leadership positions, if one is in the 30s, it's beneficial if he/she looks like and talks like he/she is in 40s, yet is fit and energetic - like Zohran Mandani. For someone in 50s, it would again be better to look like he/she is in late 40s - as that would reflect energy, wisdom and longer runway all at the same time, albeit the real age - like Shashi Tharoor was for a long time. Obama, interestingly, looked 40s, was in 40s, and behaved 40s throughout his tenure as president. True age, probably, is indeed just a number. But for leaders, there seems to be an optimal age as perceived, which may be different from the true age, and which is a function of a lot of things.

These are just my thoughts, of course, from my 'generative' intelligence on a Tuesday evening, as I struggle to make my daughter study for her test tomorrow. Not sure how I appear to her. Will know some day... or may be not🤪.

Share your thoughts.

Research - a free journey of the free mind

Research is a mindset. It’s an eternally curious state of mind that keeps walking into dark alleys, searching for something - an explanation, a deeper understanding, an inspiration, or plain nothingness that is still beautiful. It is a personal journey, through which humanity has benefited immensely.

Research that has to operate with constraints of methodology and publishability struggles to attain the depths of exploration. Having said that, methodological rigour is meant to establish robustness of findings, which is, without doubt, the most important criteria for acceptability of a research outcome. However, scholars trained to “do” research in text-book ways often fail to appreciate alternative perspectives and approaches on theory, method and robustness. This is especially troublesome in social sciences where experimental evidence is rare and theory building is supposed to be incremental and contributory.

The existing paradigms of funding and incentives around research work have created a culture of risk-aversion. Too much structure and boundedness is not healthy for scientific exploration. Research requires passion that can’t thrive under pressure to generate outcomes.

Agentic AI based Fed

What if the FOMC or the FRB was a bunch of AI agents?

It looks like a great candidate for AI. The goals and objectives of the federal reserve system are clearly laid out. The targets for monetary policy are also well established and agreed. The macroeconomic, fiscal and banking data which the fed analyses is, well, and after all, data. And AI can consume it real-time. All past research has been consumed by AI, with excellent remember and recall. Some of those near-superintelligent AI models US wants to keep for itself can be deployed here. In fact, it often seemed to me that Jay Powell was himself a superintelligent humanoid robot.

The only problem - the biggest one - would be 'independence', which is fundamental to the federal reserve system - in principle at least. While a totally human Fed has appointees pushing political agendas and personal idiosyncrasies / biases, and then there's an occasional law-suit on the Chairman, an agentic AI based system can be misled with false data. And who knows, agents may also have personalities and affiliations, depending on what went into their training. Which means for a well-intended agentic AI based Fed, there has to be a tightly controlled training mechanism, that ensures the agents are unbiased, transparent, goals-driven, meticulous, diligent and clean. But the interface with humans is where it'll all still fail, given humans will never let their control go, for both good and bad reasons. Good reasons - we can only trust humans to protect our best interests. Bad reasons - humans are divided at various levels - individual, communities based on religion, region, beliefs, experiences, economic status, affiliations, and so on; interests served vary, and not necessarily in any order or priority, and not generally good for most. What if you extend agentic AI to all governance so that Fed is not meddled directly by humans? One - AI can screw us more bigly and widely. Two - the interfaces with humans would still be there, albeit farther from the center of the agentic system, and humans at the interface can still manipulate AI to screw us. The mechanisms would be far too complex to comprehend and the outcomes far too unpredictable.

Seems humans anywhere are a problem after all. But then, who is AI for?

And...

Isn't AI 'human' too?

Fun question - What do you think an Agentic AI based Fed would have done with the policy rates yesterday?

How is human expression changing because of how LLMs express?

It's well recognized that LLMs have a certain writing style. Most of them follow common patterns like em dashes, 'a, b and c' structures, punch - then expand, 'it's x, rather than y', 'it's not p, it's q', and so on. But with so much content being written by LLMs, including that in research papers - where it's generally permitted to use AI for editing and improving readability, it's quite possible that the way humans write is also changing. With most of the content we consume being AI generated, the influence on our written expression is expected, and in my observation, starting to emerge. As evolved monkeys, most of our learning is through copying. Being quite intelligent, we are able to copy even complex patterns. The same intelligence allows us to adapt, improvise and customize - leading to both similarity and uniqueness in every aspect. (The previous line looks most definitely AI.)

I would expect the adjustment to be mutual, but I don't know whether AI is still learning from us in the way we do from it. But in future, we'll definitely have difficulty figuring out what's AI generated, because with time, NI would start looking like AI. (NI is natural intelligence. I guess I can (can I?) claim the term as the 'natural' counterpart to AI, which existed before AI.)

But the bigger concern is that, perhaps, we may not have any purely NI based writing in some distant future. We'll only be fixing drafts generated by AI based on what we tell them, just like most coders only tweak pre-written code. What we 'tell', and how we do, would be the only pure and natural expression from NI, although not polished to the eventual intended format. It's hard to imagine how that informal expression might change, but it certainly will.

There is still some way to go before we start talking to AI more than we talk to humans. That will have interesting ramifications on how we communicate, not only with AI, but also with other humans. (A close parallel, purely natural, is how American accent(s) came about, although nobody in the source countries of the OG immigrants in America spoke like that. Same goes for Australian accent.)

It's interesting how the various forms of human expression would evolve with AI playing an increasingly central role here on. I'll be watching this closely. And would love to hear what others are observing and experiencing.

IT services outsourcing - trapped in the cost play

IT services business started and evolved as a cost-reduction offering. First, offshoring offers a direct drop in costs to customers. Then greater productivity - doing more with less resources - offers greater cost savings with time. It's funny how most IT deal reviews by leaders hover around ON-OFF (onsite-offshore) ratio and YoY (year-on-year) productivity, while trying to hit an attractive price at an acceptable margin. Often, it's people with limited understanding of technology trying to buy and sell technology services.

These ratios, of course, are just high-level indicators of solution optimality, and benchmarks like 90-95% offshore headcount and 50-60% productivity-gains over 5 years are hard to push further. It's all in excel after all.

Seeking higher productivity has helped shape IT services with efficient processes, tools, and mechanisms, including automation and AI. But the pursuit is endless. Every new large deal, you are still asked - how will you bring your team to half its size in 5 years? There are no real, rather complete, answers. There's some experience, at times. There's automation. There are tools. There's AI. There's God. There's - 'we'll figure out when we get there'. Mostly, there's leap of faith compelled by fear: if we don't offer this, somebody else will, and the deal will be theirs; we have to take the risk... a calculated risk.

Peeling layers, you start looking for people mix - the pyramid. More the junior people, lower the cost - that's one lever you can pull. What about skill? - we'll train them - let's assume for now; and there's AI.

Once you've totally beaten this chain iteratively to its bare minimum cost configuration: # of people - ON-OFF ratio - Pyramid - Productivity - # of people - ON-OFF ratio - Pyramid - Productivity - ... You'd now turn to other costs - infrastructure, shifts, cabs, management, overheads, nickels, dimes, chillar!

For at least a decade now, IT service providers have recognized this need to get out of the cost play and move to value play. Mostly through talking, without really investing in genuine capabilities. So, outcomes have been only in pockets. The problem, I think, is in the nature of the buyer-supplier relationship - the relation is designed to be transactional, and both parties enter with short-term motivations.

And that probably explains why GCCs are emerging as a stronger value creation model.

The story around the picture...

Often when I look at a picture posted by someone, I can't stop myself from thinking how that moment came about, what happened before, what happened after, how that day must have looked like for people in the picture, how they interacted, what was going in each of the minds, what happened! They say 'a picture is worth a thousand words'. But the beauty of life is in what the picture doesn't convey but merely represents - all of which is in memories of those in the picture - each his/her own, fading slowly with time. And after long, the picture remains to serve only as the reminder that the moment truly happened.