Dinners, school pickups and ice cream talks have recently come down to “School WhatsApp group is too noisy” and “Elad, are we all going to die from AI?” Here’s my reply to the second most burning issue.
AI does not seem to tire from being in the headlines.
OpenAI released GPT-4o stealing the thunder from Google which made their I/O event mostly about – you guessed it –
By the time you read (assuming days from writing) new versions will be released, use cases imagined, startups dreamed, articles will discuss how the departed executives from Skynet will affect humanity’s survival. Investors will guess how Google, Nvidia, Microsoft and Meta stock price will behave. Conversations will mourn our jobs becoming obsolete and why it does not really matter since we will all be turned into paper-clips in 5 years tops.
I will try to put order, risk predictions and coin phrases to help understand the players, their strategy and the future. I will start with two spoilers: “Yes – everything is going to change” and “No – we are not going to die”.
What is the current state of AI
The AI that everyone is talking about is basically a huge data, statistics and calculation machine. It appears magical – like all good technologies when they are new – because it is hard for us to intuitively comprehend how it works. The principle is – if you feed a machine with all the word combinations ever created – it will be able to predict any new combination of words. In human terms – if you ask it anything (a combination of words) – it will be able to produce an answer (related combination of words). And here comes the “magic” – it produces the answer not because it “understands” the question – rather because it is statistically likely this combination of words is correct from the huge collection of words it is fed with.
Do the same with enough images, videos and sounds – and you get – given enough calculation power – the ability to create videos, images, sounds. And voilà – Generative AI.
How is AI going to feel like
In a year or two – AI is going to feel like speaking to a real person. Current problems will be dramatically reduced. It will stop “hallucinating” (confidently providing wrong answers), it will be cheap, it will be fast, it will be smooth. It will know a great deal of things such as medicine, math, history, philosophy, biology, physics, literature. Producing text, code, analysis, images, video and even reliably imitating “emotions”. Since using it will be like talking to an all-knowing, capable and helpful person, it will not be the domain of early adopters but everyone. Not using it will be like not using Google today. It will also affect many jobs – with painful consequences if it happens too fast – or not so dramatic if a bit slower. I will not go into predicting the detailed effect on medical doctors, lawyers, accountants, dietitians, insurance agents, support and sales reps. Let’s just say our kids will think of them like we think of scribes.
Many new things are going to emerge and “ride” the AI. Adding “spirit” to the machine. You can imagine adding an “all knowing” entity (but not intelligent – on that later) to your car? Phone? Glasses? Headphones? Tracktor? Fridge? What are you going to feel comfortable talking to it about? My guess is everything.
But there are three distinctions we need to make about AI capabilities and limitations before calling Sarah Connor and building an EMP gun. Objective AI, Subjective AI and real AGI.
The rise of Objective AI
Objective AI is (and will be much better at) giving us answers towards knowledge that is easily accessible. Anything that is somewhere on the internet or can be scanned and poured in. Knowledge, not just data, since it will be able to distill and refine according to your needs. This is the major difference between Objective AI and current search. Let’s think of a few examples –
“I have this pain in my leg, right here [showing image] above the knee. But only in the morning and only when I sit. I’m 53 and I swim regularly and never had anything like this before. Can we have a short discussion to diagnose it?”
“I recently moved to Spain and need some advice with my taxation, can you guide me through it?”
“I want to learn Japanese while commuting to work. I’m fluent in Hungarian. Can we have regular conversations, with emphasis on student life, and please correct my errors and pronunciation as we converse so I can learn from them”.
“Let’s have a discussion about virus-to-cell hydrophobic protein communication. Assume I am a postdoc in microbiology and please let’s do it in Dutch in which I am more comfortable”.
“I feel a bit off today. Can we have a chat in which I will tell you what’s troubling me and see if you can bring my spirits up?”
You get the idea. Anything that is “somewhere” on the internet or to which the companies building the AI have access to – will be available in forms which are easy for us to consume. What Objective AI cannot help with? Everything it does not have access to. Which is a lot.
Followed by Subjective AI
Subjective AI is more challenging as it requires adding to the Objective knowledge ability to access data that is not readily available. For example – what you eat and drink. Who you meet with. How you spend your time. Your financial information. What you bought at the supermarket. Your detailed medical history. And also everything that has been digitized but is behind a password. Let’s see a few examples where Subjective AI can be useful –
“Can you tell me how much I spent on toys for my kids this year and how much time they spent playing with each one?”
“Can you tell me if I’m on track to losing weight? Am I burning more calories than I consume? If so – how long is it going to take for me to reach my ideal weight and if not – can you keep guard and let me know in real time?”
“Can you call the electricity company to check why our last bill was so high, negotiate a better price and if not possible – see if you can find a competitor that will save us money per our usage profile, schedule the transfer and follow up on its completion?”
“Can you book a pottery class for me and my sister, in a studio or even with an artist that will take us as private students. It needs to be close to where we live, fit with our schedules and match our preferred style”.
These “Intelligent” conversations require access to two things –
- Data that is not currently collected. You need glasses that see what you eat and drink, what you spend time on and cameras that record everything your kids do.
- Data that is often available digitally but is behind a password. Your calendar, your Amazon purchase details, your electricity bill, your Instagram account.
OpenAI, Google, Apple, Meta. The Companion providers race
RI (Really Intelligent) as you are – patterns start emerging. OpenAI, Anthropic and other newcomers are building Objective AI. So do Google, Meta, X (Twitter), Apple. The hiccups they currently have will be solved. Errors reduced (Remember VHS? You will remember search before Larry Page invented PageRank). Prices will be lowered, interfaces smoothed, collaborations signed until everyone will have an all-knowing companion to discuss everything. Everything which is not private.
Google – will have one tiny edge over others. With your permission of course – it can access and help itself to more data. It can read your emails and documents, see your photos, know what you browse and where you are, what you buy, save your passwords and run the apps on your phone. It happens to have access to a lot of “private” data and will be happy to give more Subjective help.
To not be left behind – OpenAI can ask to use your passwords. Perhaps also building a browser and convincing Microsoft to get back to its phone-building plans. Maybe jump directly to smart glasses and simply scrap data based on everything that you do?
Apple will tell you all this is a privacy nightmare. It will provide Objective and Subjective AI right on the phone – without going to the cloud like OpenAI and Google do. It will take Apple longer (since it is hard to get the compute power required for the same level of “intelligence” to run on the phone compared to a server farm sapping the world’s energy) but when they feel confident enough – they will tell all app builders they are now obliged to open their data locally to the on-phone intelligence – providing the “Subjectivity”. Google will do the same on Android.
What about others – Meta? Microsoft? Twitter? They have access to subjective data too. Much that we are not even aware exists and is derived from usages and interactions – generating Subjectivity combined with the all-knowing Objective AI.
Hardware will be built to harness more subjective data. Watches (i.e. medical devices), glasses (combined with audio), cameras and microphones everywhere. AI Companions will be designed with different use cases in mind. The same way we do not interact with the same person for all our human needs – we will interact with different Companions for different wants.
Turing Test and the air of AGI
In 1950 Alan Turing published a paper called “Computing Machinery and Intelligence” in which he proposed a test – if a human interacts with both a machine and a human and cannot tell which is which – the machine can be deemed “intelligent”. Since then many attempts were made to achieve this but none was successful (unless employing silly tricks). It has been considered the “holy grail” of AI – a level once reached will mark the “singularity point” from which machines will start thinking independently, solve cancer, wars, poverty, aging, galactic transportation and within a few months of self-improving, enslave and then get rid of us, inferior humans.
My guess – AI will pass the Turing Test with flying colors in a year or so.
Our subsequent immediate demise – not so much.
Current AI will be incredibly knowledgeable. It will know everything humanity knows and will sound (and look) like a person. But for the same reason a knowledgeable person is not necessarily an intelligent one – so will it. AI will not have motivations, non-linear creativity, feelings, curiosity, awareness. There will be a great deal of philosophical debate about what is “intelligent” and we will be fooled by the all knowing, human seeming machines. We will feel compelled to speak with politeness and not hurt its feelings, but it will not hold a grudge. There will not be an “emergent” moment in which the machines will start improving themselves and then start conspiring against us. They will not “care”. They will not “anything”. They will be there only to answer and do the intellectual “lifting” for us.
What we did until now is “solve” language (and sound, and visual) but we do not have an “internet” source of motivation, feelings and creativity to teach a machine from. We have only the output – not the drives. These drives – which are at the core of sentient beings’ intelligence – are still out of our reach. We are still debating how to define them. Are they even “ours” (Anthropocentrism at its glory) or emerging from an ecosystem of creatures and surrounding effects operating in concert? Perhaps in the future we will be able to better understand and create a code that will “solve” this too. Then real AGI (Artificial General Intelligence) and super-intelligence might emerge.
Still worried AI will eat your lunch? No need. Unless…
Sounds dystopian? It is not. The coming AI is only another leap in efficiency.
When the printing press was invented and books were made available – regulators were terrified “reactionary ideas” will spread, breaking the social order. When trains were invented – people thought the human body traveling at speeds greater then 30 mph will break (Excusable fear. Einstein was only born 70 years later. There was no one to ask about relative speed). Calculators? Will soften kids brains. The internet? Don’t even start. Everyone will become jobless ignorants.
Things we used to spend hours, days, months and even years in order to achieve became orders of magnitude faster thanks to technology. Sure there will be “bumps” on the way (errors, privacy, fraud, addiction) and major economic shifts (whole professions will be gone) but overall – we will gain time, safety, health and wealth.
New jobs will be created and our life as a whole will be better in the long term. Not many people thought of becoming a scribe, a carriage driver or a manual loom operator in 1990. But also no one thought of becoming a Digital Marketing Manager. Not all professions will survive. If an all-knowing machine can answer better in a specific field – I would not try to study hard and compete on the ground of knowledge. “No new intelligence” also means we are still going to be very dumb as a species. A “quality” that led us to very dark places in the past.
The newly created free time can mean more creativity, more problems solved, more time spent with loved ones, in nature, doing sports. More caring for others. I’m optimistic.
One response to “Objective vs Subjective AI and The Approach of AGI”
[…] discussed in a previous post the difference between Objective AI and Subjective AI. The former gives value in relation to […]