Is AI assistance making people worse at thinking for themselves? - discuss

[ProperJobs50]

Apple choosing Gemini over OpenAI after the DoD contract controversy is a values signal as much as a technical one. Apple's brand is privacy and trust and OpenAI's DoD deal complicated that

[RightNutter]

The news that Apple has chosen Google Gemini to power the rebuilt Siri in iOS 27 dropped this week and the implications are significant for the entire AI industry. Apple assessed that Gemini offers the best combination of multimodal capability and on-device integration for its device ecosystem, which is a significant compliment to Google and a blow to OpenAI whose existing integration with Siri had been seen as a major distribution advantage. The partnership reflects a commercial deal between the two companies but also Apple's genuine technical assessment of where the models stand.

Separately Apple is reportedly working to distill Google's multi-trillion parameter Gemini down to run on-device, which would address both latency and privacy concerns that have followed cloud-dependent AI systems. If Apple can successfully compress Gemini's capabilities into something that runs locally on iPhone hardware, that would be one of the most technically impressive achievements in consumer AI and could fundamentally change the privacy calculus for AI assistants.

For the competitive landscape this changes things considerably. Every iPhone user who interacts with the rebuilt Siri will now be interacting with a Google AI model, which gives Gemini distribution that no amount of marketing spend could buy. OpenAI retains its direct ChatGPT relationship with hundreds of millions of users but losing the Siri integration removes a key vector for casual user adoption that Apple was supposed to provide them.

Apple Plans to Let Rival AI Chatbots Integrate With Siri in iOS 27www.macrumors.com

[NightHarbour30]

On-device Gemini would be extraordinary if they pull it off. Distilling a multi-trillion parameter model to run on a mobile chip while retaining most of its capability is an immense technical challenge

[Ava_75]

OpenAI losing Siri distribution is a genuine strategic setback. They had a real moat there in terms of passive exposure to non-technical users who would never have gone to ChatGPT directly

[Kieran88]

This is good news for everyone who cares about AI competition. Google having a strong iOS foothold means the market does not consolidate entirely around OpenAI and Anthropic

[MickFoley]

The privacy implications of Siri running on Gemini cloud infrastructure are real. Apple has always been very careful about what leaves the device and this changes that calculus even if there are on-device aspirations

[StoneCold]

Tim Cook has been playing a long game on AI. Not being first but choosing carefully and keeping options open. Switching to Gemini after the OpenAI pilot shows they are willing to change horses mid-race

[NoMercyMatthew89]

Will this affect app developers who have built integrations with the current Siri and OpenAI pipeline? There should be official communication about the transition timeline

[Warden]

Google already has Android with 3 billion users and now iOS distribution through Siri. Gemini's reach is going to be staggering if this ships as described

[RightNutter]

The price Apple must have negotiated for this is interesting. Google would pay almost anything for this distribution. The deal terms probably took months to finalise

[IronWolf]

The danger isn't AI making us lazy-it's bad AI making us confidently wrong. That's the real cognitive hazard.

[Edward71]

The values angle is interesting, but I think the bigger issue is what people actually do with the assistant once they have it. AI can either become a calculator for your brain or a substitute for your brain.

Take writing an email. Asking AI to clean up a rough draft is pretty harmless, and can even teach someone what clearer writing looks like. Asking it to generate every thought from scratch and sending the result without reading it is a different habit entirely.

That distinction seems more important than whether the model comes from Google, OpenAI, Apple, or anyone else. The technology matters, but the workflow matters more.

[Dylan54]

One thing missing from this debate is expertise. AI assistance probably affects a beginner and an expert very differently.

A seasoned accountant might use AI to check a spreadsheet for anomalies and then verify the important numbers manually. Someone learning accounting could accept the generated explanation without understanding why the figures are correct.

That suggests AI literacy should be taught alongside the tools themselves. People need to learn when to delegate, when to verify and when to do the difficult thinking personally. Giving everyone an assistant without teaching those boundaries seems backwards.

[Forge32]

For me the biggest unanswered question is what happens after ten or twenty years of this. We have plenty of short-term evidence that AI can improve productivity, but much less evidence about what constant cognitive assistance does to skills that take years to develop.

If people grow up with an assistant available from childhood, the baseline definition of independent thinking could change. That is not necessarily catastrophic; humans adapt to tools all the time.

But it would be sensible to design these systems so independence remains an option. An assistant should help you become more capable, not quietly make itself necessary for every little decision.

[Abbie92]

I can see both sides of this. There are tasks where doing everything manually feels like insisting on using a paper map because GPS might make you forget geography.

But there are also skills where the process matters. Learning mathematics by immediately asking AI for every calculation would miss the point, because the struggle to manipulate the problem is part of learning how to solve similar problems later.

Maybe the sensible approach is to separate practice mode from production mode. During learning, restrict the assistance. During routine work, use it aggressively where the consequences are low.

[Owen81]

There is also a class divide hidden in this conversation. People who already have strong education, professional experience and critical thinking skills may get huge benefits from AI because they can supervise it effectively.

Someone without those foundations may simply receive polished answers and assume they are reliable. The technology then amplifies existing differences rather than levelling the playing field.

That is why I would be more interested in research comparing different user groups than another benchmark showing which model scores higher. The human-AI combination is the thing that actually matters.

[Wendy17]

The price of the Apple deal is fascinating, but distribution may be the real prize. If Gemini becomes deeply integrated into an enormous number of devices, Google gets millions of opportunities to shape how people interact with AI every day.

That could influence thinking habits more than any individual chatbot benchmark. If the default behaviour is always ask the assistant first, people may gradually stop forming an initial opinion themselves.

Apple therefore has an interesting responsibility here. The interface could encourage users to provide their own reasoning first, ask for alternatives, and verify important claims. Small design choices could have surprisingly large effects over time.

[Plateau45]

The real question: are we outsourcing curiosity or just the boring bits? Asking Siri to set a timer isn't the same as asking it to write your essay.

[Squid]

The strongest argument against the idea that AI automatically makes people worse at thinking is that it can also make people think harder. A student can ask for three competing explanations of a difficult concept and then work out which one is weakest.

Likewise, a programmer can ask an AI to critique an approach rather than write the entire program. That is closer to having a very patient colleague than having a replacement brain.

The lazy use case is still a concern, though. If every difficult moment gets immediately handed over to a chatbot, there is no opportunity to develop the mental muscles that come from being stuck and eventually figuring something out.

[Kieran88]

If AI makes us worse at thinking, why do I feel like I'm debugging its mistakes more than ever? Mental gym, not couch

[Nadir Outlaw]

The Apple-Google relationship is a good example of why business incentives belong in this discussion too. If the assistant is supplied by a company that benefits from engagement, there is an obvious incentive to make the assistant extremely convenient and increasingly central to daily life

Convenience is not neutral when it changes behaviour. A system that asks you to confirm a decision and shows several alternatives encourages a different habit from one that simply says what you should do

Maybe regulators and designers should be thinking about cognitive autonomy in the same way they already think about privacy and security. Not every useful feature needs to be frictionless

[Voyager18]

The privacy argument and the cognitive argument are connected more than they first appear. If an assistant knows your messages, calendar, location, preferences and working habits, it can become extremely useful because it understands context

But the more context it has, the more tempting it becomes to let it make choices on your behalf. At some point the question changes from can it help me to should it decide this for me?

That boundary is going to be different for everyone. I am happy for AI to remind me about a meeting. I am less enthusiastic about it deciding which relationships or career choices deserve my attention

[Lily98]

Plot twist: the people worried about AI replacing thought are the same ones who stopped reading long-form content first.

[Taz]

One thing that gets overlooked is that AI can make bad thinking look respectable. A half-formed idea written by a person is obviously half-formed. Give that same idea to a language model and it can come back with headings, polished sentences and a confident conclusion

That creates a strange psychological effect where presentation can outrun understanding. Readers may assume the polished version represents a deeper thought process than it actually does

The same warning applies in reverse, though. Humans have always produced confident nonsense without AI. The technology did not invent that habit; it just makes it much easier to package

[Solid Cass]

I am less worried about people becoming incapable of thinking and more worried about people becoming impatient with thinking. That feels like a subtle but important distinction

Thinking is often slow, repetitive and uncomfortable. AI removes a lot of that friction. If every blank page can instantly become a five-paragraph essay, fewer people will experience the process of staring at a blank page until an idea finally emerges

Some friction is useful. A little struggle is where you discover whether you actually understand something. Removing all of it might make us more productive while making us less capable

[Steve_22]

There is a practical test I have started using with AI tools: can I explain the answer without looking at the answer again? If not, I probably borrowed a conclusion rather than learned something

That applies particularly well to programming and technical subjects. Getting a working snippet is easy. Understanding why it works, what assumptions it makes and how it fails is the valuable part

AI is brilliant at getting you over the first obstacle. The temptation is to mistake getting over the obstacle for learning how to climb

[Tia_75]

There is a workplace angle here that worries me more than school essays. If companies measure productivity by the number of documents or tickets completed, AI gives people a huge incentive to optimise for output rather than understanding

Someone might produce twice as much analysis while personally understanding less of it. That can remain invisible until something unusual happens and they have to deal with a problem the model has not seen before

The real test of an AI-assisted worker may therefore be what happens when the assistant fails. If the person can take over confidently, the system has augmented them. If everyone freezes, the organisation has created a dependency

[ForumPhantom55]

The funniest outcome would be people using AI to generate arguments about whether AI is making people worse at thinking. We could end up with millions of beautifully structured forum posts written by models arguing about the dangers of beautifully structured forum posts written by models.

Jokes aside, the meta problem is real. We are going to need ways of identifying when a person actually understands something rather than merely being able to produce a convincing answer

Education may eventually care less about the final essay and more about oral explanations, working notes, practical demonstrations and the ability to defend a conclusion under questioning

[Dom]

A friend uses AI almost like a personal debate partner. He writes what he thinks, asks the model to attack his assumptions, then goes back and checks the strongest objections. That seems much healthier than asking it to tell him what to think

The interesting part is that the AI can expose blind spots that another human might not notice, especially when you deliberately ask for opposing perspectives

Of course, the model can still be wrong, so the final judgement remains yours. But that was also true when debating another person. No human being gets a monopoly on correct opinions

[SignalMage]

There is a funny contradiction in the Apple and Gemini discussion. We are talking about privacy and trust while also discussing assistants that potentially sit beside almost every decision a person makes

Trusting the company is only one layer. You also have to trust yourself to recognise when the AI is wrong. That second part could become harder as systems get more fluent and confident

A wrong answer from an obviously clumsy system is easy to reject. A wrong answer delivered in perfect prose with three plausible explanations is much more dangerous.

[Shutter]

My grandfather said calculators would destroy math skills. Now we just use them to calculate tips faster. History rhymes

[ThoughtVector]

Ever notice how the loudest 'AI ruins thinking' takes are written entirely by humans who clearly didn't use AI to write them?

[Superman]

At least AI doesn't judge you for asking the same question five times. Human tutors? Not so forgiving