LinkedInDeterministic automation is not the enemy of intelligence. It is the floor, the rockbase.. Without determinism, you cannot build trust. Without trust, you cannot deploy autonomy. And without autonomy grounded in real dynamics, all the tokens in the world will never get you 'intelligence'. So the real question for 2026 is not whether AI goes boom or bust.. It is whether we stay trapped in probabilistic or deep learning dogma, or finally move toward systems that can adapt, act, and remain coherent in the real world, in realtime, under uncertainty.. History suggests that when the hype cycle collapses, the real work finally begins.
For me, history suggests that without proper root cause analysis, we are doomed to repeat the same story. The question is why this GenAI hysteria could even emerge while the science and engineering was always clear, unambiguous and known? How could this misinterpretation of the Turing test become a global cargo cult? How could this whole industry not even reject but totally forget its own foundations?
Todd Johnson
Lorand Kedves For me, it is because the coding agent (in particular Claude and Claude Code) is remarkable. I can finally do things that I did not have the resources to do. I have been at universities since 1980. We are always underfunded--even with grant funding, which never provides enough to do the work that needs to be done (much less the extra committee, teaching and grant writing work). At this stage, it is a matter of learning when and how to use the LLMs, not whether to use them.
Thoralf J. Klattit’s not just the efficiency gains. also innovation that comes from understanding hashtag#unmetneeds. we don’t just need to do the wrong stuff faster but understand what’s the right stuff. solving people‘s needs (aka pro•duct in latin). enter jobs-to-be-done. Eckhart Boehme can tell you how to augment with hashtag#graphRAG to make more sense. start here:https://medium.com/@thoralfjklatt/better-discovery-using-jobs-to-be-done-jtbd-46775f2a2234
Eckhart BoehmeThoralf J. Klatt A highly controlled process for building a data basis (human behavior) for context-based graphrag produced by deterministic and probabilistic rules and a step-by-step process is a middle way to produce meaningful results.
Lorand KedvesThoralf J. Klatt - sounds like "I think the computer has from the beginning been a fundamentally conservative force. It has made possible the saving of institutions pretty much as they were, which otherwise might have had to be changed." (Joseph Weizenbaum, 1985) 🤔
https://lnkd.in/dPmbG4TN
Eckhart Boehme - I guess JTBD would not be new for the people of the Apollo program or Skunkworks...
https://youtu.be/ecIWPzGEbFc?t=2851Or a "context-based graphrag produced by deterministic and probabilistic rules and a step-by-step process" to Ivan Sutherland (1963...)
https://youtu.be/6orsmFndx_o?t=45This is what I call the forgotten foundations, with the predicted consequences... 👇
More on that in my lame summaries
in text:
https://lnkd.in/d8_cAMBBin video:
https://lnkd.in/duJEbZty
Todd Johnson - speaking of academy.
1: I have no idea what an LLM can do with the formal definition of the term information: "things that you did not know and could not derive from what you knew" vs taming gigantic neural nets with examples. 🤦♂️
2: I know that Turing's article was not a goal but a warning, and his real challenge was to define the terms 'machine' and 'thinking' BEFORE making any statements with them. 👇
3: I know that Noam Chomsky's research on languages is the theoretical and practical foundation of modern programming, lately realised that his generative grammar is the key to answer Turing's call.
https://lnkd.in/dTCD8dBvIt is incomplete, but not "crazy" as proclaimed by "a psychologist trying to understand how the brain works" a.k.a. "the godfather of AI"... 🤷♂️.
https://youtu.be/aAvtBdtyEOgGambling is a known human weakness exploited by huge industries. We can spend trillions on hyper-sophisticated slot machines and call that 'rocket science'. Unfortunately, true rocket scientist still have to eat and pay their bills, so the field of rocket science disappears.
https://youtu.be/Elyfo1DIlzs?t=91Examples of that true rocket science:
https://youtu.be/_OSspHZICOghttps://youtu.be/aXVUoT_objA
Thoralf J. Klatt Lorand Kedves it’s about increasing the likelihood that customers make progress by adopting your solutions. JTBD originates in marketing which turned out to be part of your pro•duct
Lorand KedvesThoralf J. Klatt As I know, Turing assumed the total understanding audience of his real technical articles around ten (10) on the planet. This is how you do real science. The success of going to orbit is not about marketing or consumer adoption. It is about physics. This is how you do real engineering.
Don't get me wrong, you are right in the business and unfortunately, society and academy as well (Aaron Swartz learned this the hard way
https://youtu.be/9vz06QO3UkQ )
I am afraid we disagree on the definition of "real work"...
One more thing... for you and the maybe 10 people still following this thread 🤫
You set two goals,
1: "we don’t just need to do the wrong stuff faster but understand what’s the right stuff"
2: "customers make progress by adopting your solutions"
They are in contradiction!
Every system has an inertia, customers want to make progress in their current path and mental framework. Any solution that would require additional thinking, learning or new tooling is rejected because of the decreasing immediate profit and risk. This is the only thing these people would agree on:
Alan Kay -
https://youtu.be/NdSD07U5uBs?t=787Ted Nelson -
https://youtu.be/KdnGPQaICjkDouglas Engelbart started investigating this problem in his research of AI (as Augmenting Intellect), Dynamic Knowledge Repository, NLS, Boosting Collective IQ, ...
https://lnkd.in/dsEEzZVs His A-B-C model resolves that contradiction by layering.
A: improve a current process.
B: analyse the 'A' solution, find ways to change anything that would bring improvement in the long run.
C: consider 'B' as 'A', improve the improvement process. This is where we are get back to "crazy" Chomsky again, against the whole genAI tragicomedy.
Thoralf J. KlattLorand Kedves the 4 forces of JTBD are in balance until customers pull, habits are overvome and anxieties vanish. that’s the moment of your pro•duct leading forth to solution of their needs. they switch. see this video by Bobby Moesta incl a customer interview from Toronto. AI can help humans in discovery and understanding experience design when made available in graphs. https://vimeo.com/81153746
Eckhart BoehmeLorand Kedves the principles didn't change but the tools.
Lorand KedvesEckhart Boehme Yep... knowing some "old tools" and being aware of the hardware capabilities, the comparison is disappointing.
Here is Alan Kay from 2003
https://youtu.be/1pXmuh1AUQQ?t=4933Or Engelbart's NLS (1968, before the test launch of ARPANET).
https://youtu.be/UhpTiWyVa6kThe hardware difference from the same Bob Martin lecture
https://youtu.be/ecIWPzGEbFc?t=3142I don't think "modern tools" reflect the elapsed 20+ or 50+ years. 🤔
Thoralf J. Klatt "AI can help humans in discovery and understanding experience design when made available in graphs." - exactly, if we trash chatbots and return to what Ivan Sutherland worked on in 1963.
"- We're going to show you a man actually talking to a computer in a way far different than it's ever been possible to do before.
- Surely not with his voice.
- No he's going to be talking graphically. He's going to be drawing and the computer is going to understand his drawings. The man will be using a graphical language that we call Sketchpad."
https://youtu.be/6orsmFndx_oBTW, this was the topic of my CS PhD in 2018 (age 45). Rejected of course, they "did not see the academic value"... 🤣 Here is a very obsolete prototype from that time:
https://youtu.be/3GsSp7Zd1g8
Thoralf J. KlattLorand Kedves the open questions to be asked can be designed by AI. humans still need to understand and make sense of the needs of humans. after all they will buy your product. robots will only buy energy and water
Lorand KedvesThoralf J. Klatt ... and we are back at the beginning: how does 'an AI design a question'? What is 'a machine' and what is 'thinking'? 🙂 / 🧠 / 🤖
For some seasoning, add Fred Hoyle's Black Cloud (if an intelligent entity should be human-like), or history / current politics (if being human-like actually means intelligent). 😁
Thoralf J. Klatt Lorand Kedves based on the 4 forces (push, pull, habits, anxieties) and functional, emotional personal, emotional social, life-changing, financial (unmet) needs. The answers are the interesting part. As always. AI is your personal assistant hashtag#PA, nothing more. You are accountable as a product owner. Hope this helps. Try it. cc Riccardo Mariti
Todd JohnsonWhen I use LLMs for coding, I'm not gambling or expecting it to pass the Turing test... I'm getting work done--more work than I ever got done before. My concern is that outside of their training distribution and outside of the custom agentic systems (such as Claude Code) they show very limited ability to reason even in areas where humans can do quite well. They are still stochastic parrots, albeit more and more clever ones and more and more useful ones. But leadership in most cases is bound to be fooled by these parrots, because they don't understand the limits and will be all too willing to accept the hype.
Lorand KedvesTodd Johnson I know, but this is the problem, not the solution (painkiller vs cure).
I am an IT expert writing code for 40 years, got CS BSc in '99 (unlike the famous hype riders, and not in psychology, like prof. Hinton). I was a lead developer of an AI startup 25 years ago, we won Comdex '99 with DataScope, a data visualisation and knowledge extraction tool supported by "traditional" machine learning. I improved my knowledge graph tooling behind many projects (government, multinationals, startups, academic R&D) since then. Meanwhile I got suspicious that I missed something, went BACK to the academy at age 43 for a CS MSc and a half PhD (they rejected my research).
You are a very experienced and tech-aware client, one that I would love to work with. Now you have the illusion that you can accomplish more with your chatbots than ever before. But whatever you do is not more than the rough average of all previous works, that's the content of the LLM. Congratulations, you hired a million script kiddies. 🎉
But they will never have my knowledge because that is not the average but the exception. And nobody cares if I solve problems, because the money is NOT in solving but 'dealing with' them.
https://lnkd.in/dy5Tnm3nCatch-22.
