Some of my thoughts on AI

29 Apr 2023 | in ai

Translations: en en

29 Oct 2022

Key traits in an AI world:
- adaptability
- system thinking
- human connection

20 Jan 2023

Extrapolate progress in generative AI & brain interfaces to a not so far future, and we should be able to visualise & share recordings of our dreams and visual thoughts.

This will unlock a new level of human creativity & communication possibilities.

31 Jan 2023

The threat of population collapse is counter-intuitive, as we have now reached 8 billion people on Earth.
Yet last week alone, Japan & Italy have raised the alarm that their countries might not be able to function, and their health system fail, if the birth rate continues to trend down.

This might be a temporary (decades long) issue, if new policies & societal changes help get this trend back up.

But the more I think about it, the more I think AI can be part of the solution.

Instead of fearing it taking jobs away, it can help empower a dwindling active population to be more productive, thus helping supporting economically an ever-growing aging population.

It's a complex topic though.

Has anyone looked into this topic in depth?

Source of graph: World Bank


01 Feb 2023

A few tools (incl. from OpenAI) are in the works to help detect if something has been created by an AI or a human.

I think broadly an AI-detecting AI is a short-term, knee-jerk reaction to a new technology society fears and will need to adapt to.

Everyone will use AI assistants in the future.

And as we have started to see, two different people using the same AI will get different results. Knowing what & how to ask the AI is a skill in itself.

But if you take school curriculums for example, they will need to evolve anyway - there has been a disconnect between what is taught (and how) and what people need in the workplace & life, for a long time already.
The gap is just widening exponentially now.
It's about rethinking what and how we teach, rather than keeping the status quo and fighting the rise of AI.

Plus there are risks of false positives and false negatives.

Sure there are use cases where an AI-detecting AI will have a role, eg to combat the spread of misinformation and deepfakes used for nefarious purposes.

But ultimately what matters in most cases is output quality - not how the output was created.

Do we really care how something good has been created?

Given the choice between something sub-par created 100% by a human or something better created partly or fully by an AI, which would you choose?

Give it a few years and "can I speak with your AI?" might become the new "can I speak with your supervisor?".

Will AI replace jobs?

In some cases, yes, but mostly:
1) it will replace people not good at the job or with low human added-value
2) it will enhance & empower people good at the job, who embrace AI
3) it will create new jobs

NFTs might see a revival in the years ahead (under a new name I hope) as a way to prove that a piece of content was created by a human.

AI #irony

Carl Wilhelm Hagander 🦆 I think about the human interaction factor in the rise of AI, and I'm split on that one...

(disclaimer: I'm an introvert who likes efficiency, so probably on the wrong side of the argument :)

My current thinking, in a world where AI output is trusted to be better and more consistent than humans, is:

  • in most cases, in a business setting, once the digital-native generations represent the largest majority of white collars, an efficient AI interaction will be preferred (ie do more in less time, with a higher certainty of quality output)
  • this in turn will augment the need/desire for authentic human interactions in one's personal life

Obviously, I might be wrong 🤷🏻‍♂️

Joey Gilkey Agree with your underlying message. Though it's about focusing on signal vs noise.

And extrapolating what (generative) AI will be able to do in the years to come - always gotta learn & adapt.

The flipside question would be: If you were good at X before, how much better can you/will you be with AI assistants?

I think the evolution of wealth distribution is a complex and dynamic process - hard to argue in 300 characters.

Plus I'm a "glass half full" kinda guy :) Doesn't matter if a small number of people get richer, as long as overall wealth distribution gets better in the process 🤷🏻‍♂️

Also the flipside I've been thinking about recently - with Italy & Japan raising alarm over falling birth rates and upcoming potential collapse of their society & health systems - is the promise of AI to empower a dwindling number of active workers to be more effective/productive, so as to be able to support an ever-growing aging population. 🤔

08 Feb 2023

A few years from now, an AI should sit at every meeting table.


  • take notes / transcribe
  • send summary & action points to attendees
    but also:
  • fact-check based on both internal & external data sources
  • provide an additional opinion
  • help brainstorms ideas
  • keep the meeting on track / ensure goals are met
  • assess participation and raise flags if underrepresentation
  • suggest ways to improve meeting efficiency

Love this. A good counter-balance to the overall backlash against AI's inaccuracies.. as if humans & mainstream media would never be wrong 🤷🏻‍♂️

That being said, generative AI needs to improve with:

1) correcting the "hallucinations" phenomenon
2) hyperlink to sources
3) indicate confidence score

Early days still.. 5 years from now these teething issues will be forgotten.

11 Feb 2023

(Human + AI) > (Human or AI)

Human + AI will outperform Human or AI for a long time.

14 Feb 2023

A few years ago, experts were saying that creativity will remain the remit of humans for a long time. Proven wrong very quickly.

When thinking now about what will remain a "human" trait in this new dawn of AI, curiosity (ie asking good questions) came to my mind.

It's what has propelled humanity to explore, discover & evolve.

But developing a curious AI is probably around the corner too 🤷🏻‍♂️

Re Sales, I think the impact, at scale, will take 5 years rather than 2. But it will be profound.

True value will be unlocked once any company is able to easily train its own large language model on its own data, best performers' playbook, target industries, etc..

Combined with progress in other areas (eg automation), my best guess is that you'll be able to get the same output with 60% less salespeople + AI.

Fascinating times ahead.

29 Mar 2023

Yes. It's going too fast at the moment, without railguards. I'm excited by the progress, understand the underlying technology, and don't fear any sentient AI. "Hallucinations" in text replies are a non-issue. But danger lies in:
- ideological bias (this is a big step from solely commercial bias, in a democracy)
- powerful & easy deepfake video & audio generation accessible to all
- most importantly: easy access to AI generating & executing code

28 Apr 2023

AI dangers

  • ability to manipulate humans with better-than-human persuasion, based on the entire web & social media data.
  • trained on lies as well as truths, with no moral compass.
  • agents bring a whole other level of autonomy & risk.

AI / Yann LeCun

I don't think anyone says "stupid enough" - maybe just "not intelligent enough" as a human species considering the speed at which super-powerful AI can develop itself at some point and the ramifications for good and bad it creates.

Just like Facebook did not predict the negative consequences of its algorithm, on society & people.

Could lead to have unintentionally harmful consequences at a much larger scale than the first few cars and planes you take as example.

And we have learned with cars and planes that regulation is needed to ensure broad safety.

I respect your decades of experience & knowledge, but AI timeline & expected capabilities have been mis-forecasted before by many experts.

You might/will be not at fault yourself if a powerful AI goes rogue - it might/will be someone else. It's about protection across the board.

Let's just have at least a regulation around implementations of hardware & software fail-safes.