See All AI Articles
Picture a sunny Madrid evening, wine glasses clinking, and three Nobel laureates arguing about whether your next best friend—or boss—will be a machine. That was the Nobel Prize Conversation on “Our Future with AI,” streamed live from the Fundación Ramón Areces. Here’s the cheat-sheet for normal people.
- Will AI steal creativity?
Short answer: It’s already borrowing it, but it’s not wearing the T-shirt yet.
- Geoffrey Hinton (the “godfather of deep learning”) showed how ChatGPT spotted that a compost heap and an atom bomb are both slow vs. fast chain reactions—an analogy most humans miss.
- Serge Haroche (Nobel for trapping single atoms) says real creativity is driven by goose-bump curiosity, not pattern-matching.
- The compromise: AI is like a super-fast intern who brings you wild ideas, but you still decide which ones are worth building.
- Quantum vs. AI: the next “weird science” team-up
Imagine atoms as tiny magnets that can talk to each other “in secret.” When too many join the conversation, even Einstein got dizzy.
Haroche’s lab now uses AI to spot patterns in that quantum chatter, cutting experiment time from weeks to hours. Translation: faster quantum computers, better drug design, weirder gadgets for your living room.
Haroche’s lab now uses AI to spot patterns in that quantum chatter, cutting experiment time from weeks to hours. Translation: faster quantum computers, better drug design, weirder gadgets for your living room.
- Can AI crack my passwords?
Not in a clever new way—yet.
María Isabel González Vasco, a crypto-math wizard, says AI just speeds up old tricks (like listening to the faint whirr of your laptop to guess your key). Her advice: keep updating your software; the really scary stuff is still quantum computers, not ChatGPT.
María Isabel González Vasco, a crypto-math wizard, says AI just speeds up old tricks (like listening to the faint whirr of your laptop to guess your key). Her advice: keep updating your software; the really scary stuff is still quantum computers, not ChatGPT.
- Robots in the classroom: tutor or terminator?
Hinton likes the idea of an AI tutor that never gets tired, adapts to every kid, and lets human teachers do the mentoring.
Haroche worries we’ll forget how to think if we outsource too much.
Consensus: blended families work—blended classrooms might too, as long as we pay teachers like we pay hedge-fund managers (spoiler: we don’t).
Haroche worries we’ll forget how to think if we outsource too much.
Consensus: blended families work—blended classrooms might too, as long as we pay teachers like we pay hedge-fund managers (spoiler: we don’t).
- Jobs: which ones first?
Elastic jobs (healthcare, creative gigs) will expand—AI makes a nurse or designer 10× more productive, and we’ll simply want more care and more stories.
Elasticity-lacking jobs (call-center scripts, box-ticking) are toast.
Political punchline: productivity itself isn’t the enemy; who pockets the profit is.
Elasticity-lacking jobs (call-center scripts, box-ticking) are toast.
Political punchline: productivity itself isn’t the enemy; who pockets the profit is.
- How do we stop SkyNet?
- Tech fix? “Mechanistic interpretability” (think MRI for software) helps, but it’s early days.
- People fix? Public pressure, same playbook as climate change: write to your politician, join a citizens’ panel, ask “Who’s liable when the algorithm goofs?”
- Europe’s ace card: it’s a 450-million-person market—if Brussels demands “bias labels” or “energy passports” for AI, giants like Google have to listen.
- Should I panic?
Hinton’s gut: 10–20 % chance of really bad outcome (think sci-fi level).
Haroche’s gut: civilisation is driving toward several walls at once—climate, nukes, AI—but giving up science is the one guaranteed crash.
Practical takeaway: worry less about killer robots, more about dull ones that deny your mortgage because your postcode looks “risky” to the training data.
Haroche’s gut: civilisation is driving toward several walls at once—climate, nukes, AI—but giving up science is the one guaranteed crash.
Practical takeaway: worry less about killer robots, more about dull ones that deny your mortgage because your postcode looks “risky” to the training data.
- Three things you can do this week
- Treat AI like a powerful stranger: great for restaurant tips, terrible for secrets. Don’t feed it your medical records.
- Ask your kid’s school how they use AI tutors—push for transparency.
- Follow one AI-safety newsletter (e.g., AI Policy Weekly)—five minutes a week beats doom-scrolling.
- The “black-box” problem: why explaining AI is harder than building it
Imagine your sat-nav sends you down a goat path instead of the motorway. You can open the app and see why: road-works, accident, faster-time route.
Now imagine an AI denies your loan. The bank shrugs: “The computer says no.” That’s the black-box problem.
Now imagine an AI denies your loan. The bank shrugs: “The computer says no.” That’s the black-box problem.
What the laureates told us
- Geoffrey Hinton is betting on “mechanistic interpretability” — basically giving the network an X-ray. Early results show individual neurons lighting up for concepts like “legal” or “DNA.”
- María Isabel González Vasco warns that even if we spot bias, fixing it is like playing Whack-a-Mole. Delete one unfair signal and the model finds a proxy (zip code, browser type, even the font you used on your CV).
- Serge Haroche’s physics analogy: in quantum mechanics we can’t see an electron directly, but we can measure its shadow in a cavity. Same trick is now being tried on neural nets: watch how they change a story when you swap a name from “Emily” to “Emilio.”
Practical takeaway for non-coders
If a company can’t explain an AI decision to you in your language, treat it like a used car whose engine you’re not allowed to open — walk away, or at least ask for a warranty.
If a company can’t explain an AI decision to you in your language, treat it like a used car whose engine you’re not allowed to open — walk away, or at least ask for a warranty.
- Energy: the hidden bill we’re not paying
Training a big model uses roughly the same electricity as 5 000 households in a year. And that’s before anyone starts chatting with it.
Why the brain still wins
Your skull runs on 20 watts — less than a fridge light. A super-computer matching one human brain needs a million times more power. Haroche jokes that evolution had a 3.5-billion-year head-start and a strict electricity budget.
Your skull runs on 20 watts — less than a fridge light. A super-computer matching one human brain needs a million times more power. Haroche jokes that evolution had a 3.5-billion-year head-start and a strict electricity budget.
What can be done
- Green data centres: Google and Microsoft already buy wind and solar to match annual consumption, but the grids still go brown on calm nights.
- Tiny specialised chips: your phone’s AI camera runs on a sliver of silicon that does one job brilliantly. Expect more of those in hospitals, cars, even coffee machines.
- Algorithmic thrift: new “pruning” methods literally snip away 90 % of the network after training, like editing a 200-page draft down to 20 without losing the plot.
Personal angle
Next time you ask ChatGPT to write a limerick, you won’t crash the planet. But if you’re a company running millions of queries an hour, the kilowatt-hours stack up fast — and investors are starting to ask why your electricity bill just overtook your coffee budget.
Next time you ask ChatGPT to write a limerick, you won’t crash the planet. But if you’re a company running millions of queries an hour, the kilowatt-hours stack up fast — and investors are starting to ask why your electricity bill just overtook your coffee budget.
- The privacy swamp: your data is the new oil, but who owns the well?
María Isabel shared a classroom experiment: she asked 30 computer-science students to use an AI résumé tool. Within 24 hours 12 had uploaded their full medical histories “to get better wording.” None read the terms-and-conditions.
Three creepy truths
- Anything you type can, and probably will, be used to train the next model.
- Even “anonymised” text leaks: postcode + rare hobby + pet name often re-identifies you.
- Deletion requests are voluntary outside the EU, and inside the EU they can take up to 30 days — long enough for your data to be baked into a trillion-weight cake.
Simple self-defence kit
- Use the browser’s “private” mode when you experiment.
- Strip names, addresses and numbers before you paste text.
- If you’re an employer, add a clause: “Staff must not feed proprietary data into public AI tools without approval.” (Most Fortune-500 boards still haven’t done this.)
- Creativity remix — can AI be original?
The compost-heap test
Hinton’s favourite party trick is asking, “Why is a compost heap like an atom bomb?” Most people stare blankly. GPT-4 answered: both are chain reactions where heat speeds up the process, just on different time-scales. That’s not in any textbook — it emerged from the model squeezing knowledge into fewer connections.
Hinton’s favourite party trick is asking, “Why is a compost heap like an atom bomb?” Most people stare blankly. GPT-4 answered: both are chain reactions where heat speeds up the process, just on different time-scales. That’s not in any textbook — it emerged from the model squeezing knowledge into fewer connections.
But is that real creativity?
Haroche says no — because the network never felt the aha! moment. It didn’t risk tenure, lose sleep or jump up and down when the analogy clicked.
Philosopher’s corner: if creativity is defined as “seeing connections that matter to us,” then humans still hold the steering wheel. If it’s just “produce something statistically novel,” AI is already a prolific artist.
Haroche says no — because the network never felt the aha! moment. It didn’t risk tenure, lose sleep or jump up and down when the analogy clicked.
Philosopher’s corner: if creativity is defined as “seeing connections that matter to us,” then humans still hold the steering wheel. If it’s just “produce something statistically novel,” AI is already a prolific artist.
Try it yourself
Ask your favourite chatbot to invent a sport that could be played on Mars. Then ask it to invent the rulebook, equipment list, and safety waiver. You’ll get pages of plausible text in seconds. Now try to play the game with friends. Suddenly you’ll discover the gaps only a human body — and human humour — can spot.
Ask your favourite chatbot to invent a sport that could be played on Mars. Then ask it to invent the rulebook, equipment list, and safety waiver. You’ll get pages of plausible text in seconds. Now try to play the game with friends. Suddenly you’ll discover the gaps only a human body — and human humour — can spot.
- Jobs part 2: the ones you hadn’t thought of losing
We expect taxi drivers and call-centre staff to be squeezed, but the Madrid panel flagged some surprises:
- Junior lawyers: discovery work (sifting millions of emails) is now 80 % faster with AI.
- Radiographers: AI spots lung nodules better than a first-year resident; the human role shifts to comforting patients and double-checking edge cases.
- Voice-over actors: your audiobook can be read in your favourite celebrity’s cloned voice for a few hundred dollars.
- Code-copyists: developers who mainly glue Stack-Overflow snippets together are discovering the AI does that in milliseconds.
The flip side
New gigs are popping up: prompt engineer, model auditor, AI-ethics trainer, synthetic-data curator, “human-in-the-loop” storyteller. None existed on LinkedIn five years ago; today they’re six-figure niches.
New gigs are popping up: prompt engineer, model auditor, AI-ethics trainer, synthetic-data curator, “human-in-the-loop” storyteller. None existed on LinkedIn five years ago; today they’re six-figure niches.
Career advice nobody asked for
Move upstream: ask why the code, the image, or the diagnosis is needed in the first place. Machines are brilliant at how; humans still own why.
Move upstream: ask why the code, the image, or the diagnosis is needed in the first place. Machines are brilliant at how; humans still own why.
- The geopolitical chessboard
USA vs. China vs. Europe in one slide
- USA: piles of venture cash, relaxed rules, “move fast, regulate later.”
- China: gigantic data sets (1.4 billion faces), state-backed fusion of surveillance + commerce.
- Europe: no tech giants but 450 million affluent consumers → uses market size to write the rulebook (see GDPR, now copied worldwide).
What the laureates want
Haroche: “Europe should play referee, not just striker.”
Hinton: “If democracies don’t coordinate, authoritarians will set the default settings for everyone.”
Vasco: “Privacy standards born in Brussels end up in Buenos Aires and Bangalore — let’s make them good.”
Haroche: “Europe should play referee, not just striker.”
Hinton: “If democracies don’t coordinate, authoritarians will set the default settings for everyone.”
Vasco: “Privacy standards born in Brussels end up in Buenos Aires and Bangalore — let’s make them good.”
Ordinary-person leverage
Every time you choose a product that boasts “GDPR-grade privacy” or “EU AI-Act compliant,” you cast a vote for that rulebook. Companies track those votes with the same fervour they track clicks.
Every time you choose a product that boasts “GDPR-grade privacy” or “EU AI-Act compliant,” you cast a vote for that rulebook. Companies track those votes with the same fervour they track clicks.
- A day in your life, 2030 edition
07:00 — Your AI alarm composes a wake-up song based on your heart-rate data.
07:30 — Coffee machine brews a new blend invented overnight by a generative model trained on your past ratings.
08:15 — Autonomous bus reroutes around a street fair you didn’t know was happening; you read the summary aloud in Spanish even though you never studied it — real-time translation earbuds.
09:00 — Doctor’s visit: an AI has already ruled out 12 rare diseases, so the human physician spends the full 15 minutes discussing your anxiety about them.
12:30 — You lunch at a pop-up restaurant whose menu was created by AI to use only leftovers from yesterday’s food-delivery surplus.
14:00 — Work: you spend two hours “mentoring” an AI through ethical edge cases; your signature is required before it can release the new drug-recipe to regulators.
18:30 — You play a VR board-game set on Mars; the storyline adapts to your kid’s homework on volcanoes.
22:00 — Bedtime: the room lights dim in a pattern proven (on people like you) to maximise deep sleep, but you can still override with one tap.
07:30 — Coffee machine brews a new blend invented overnight by a generative model trained on your past ratings.
08:15 — Autonomous bus reroutes around a street fair you didn’t know was happening; you read the summary aloud in Spanish even though you never studied it — real-time translation earbuds.
09:00 — Doctor’s visit: an AI has already ruled out 12 rare diseases, so the human physician spends the full 15 minutes discussing your anxiety about them.
12:30 — You lunch at a pop-up restaurant whose menu was created by AI to use only leftovers from yesterday’s food-delivery surplus.
14:00 — Work: you spend two hours “mentoring” an AI through ethical edge cases; your signature is required before it can release the new drug-recipe to regulators.
18:30 — You play a VR board-game set on Mars; the storyline adapts to your kid’s homework on volcanoes.
22:00 — Bedtime: the room lights dim in a pattern proven (on people like you) to maximise deep sleep, but you can still override with one tap.
Creepy or cool? The difference is whether you can read the settings menu — and switch features off.
- TL;DR cheat-sheet to sound smart at dinner
- AI is already creative — but only inside the playground we build.
- Quantum + AI = faster gadgets, not magic wands (yet).
- Your passwords are safer from AI than from your own reuse of “Fluffy2020.”
- Teachers aren’t doomed; lecture-style teaching is.
- Energy use is the silent crisis — efficiency matters more than size.
- Europe’s super-power is standards, not servers.
- You’re not helpless: demand explanations, read menus, bug your MP, and never upload your medical file to a chatbot.
Closing thought
As one laureate put it, “Creativity is connecting two dots that were always there, but nobody had bothered to draw the line.”
AI just handed us a bigger box of crayons. The picture we draw is still up to us.
AI just handed us a bigger box of crayons. The picture we draw is still up to us.
No comments:
Post a Comment