Could decentralization in AI make a difference ?

[Firstfrost]

In this space, we’ve always known centralization as an obstacle, we’ve seen it in banking industries, exchanges and some other sectors. Now AI seem to be following the same path with a few companies controlling the computing power and access. Looking like those days before decentralization.

Lately I’ve been curious about how AI run across different systems instead of a single company’s servers. Realized we have DecentralGPT now and i find the idea quite interesting. It’s aimed at making AI cheaper, more accessible and more private.

But what do you think about this ? Will AI becoming decentralized actually make a difference ?

[Ojima-ojo]

It can't operate on a Decentralized model as long as AI requires human input to function as designed, so there will always be interference, this is the limitations that comes with such development, unlike Bitcoin and the blockchain that designed a well structured decentralized network that allows users to choose between centralization and Decentralized self control power, this made a lot of difference between the two development.

[passwordnow]

In this space, we’ve always known centralization as an obstacle, we’ve seen it in banking industries, exchanges and some other sectors. Now AI seem to be following the same path with a few companies controlling the computing power and access. Looking like those days before decentralization.

Lately I’ve been curious about how AI run across different systems instead of a single company’s servers. Realized we have DecentralGPT now and i find the idea quite interesting. It’s aimed at making AI cheaper, more accessible and more private.

But what do you think about this ? Will AI becoming decentralized actually make a difference ?

If a company needs to be operated and maintained well, there can be help from an AI to do all of the automation that server has to be. But I don't think that a company is willing to do that without having human intervention. If they have an important affairs with their company, it is always needed to have someone have it monitored and still do a human job for their own sake. Not because the AI have been into the trend lately and no doubt that it's helping a lot of companies to be more productive, I am sure that many of them won't go that further unless proven already that it can work on its own and if it's decentralized, I don't think that they'll like it more.

[Iamcrypticguy]

In this space, we’ve always known centralization as an obstacle, we’ve seen it in banking industries, exchanges and some other sectors. Now AI seem to be following the same path with a few companies controlling the computing power and access. Looking like those days before decentralization.

Lately I’ve been curious about how AI run across different systems instead of a single company’s servers. Realized we have DecentralGPT now and i find the idea quite interesting. It’s aimed at making AI cheaper, more accessible and more private.

But what do you think about this ? Will AI becoming decentralized actually make a difference ?

Anything decentralized makes a difference cos it gives room for competition, which may improve performance!. I'm not really into AI though except when I use GetAgent to analyze trades. Decided to check decentralGPT you mentioned and discovered something interesting, their token will debut on multiple exchanges laster today include Bitget, kucoin etc.. with that much pull, they must be doing soemthing right.

[TastyChillySauce00]

Sometimes I wondered about these kind of "decentralized AI" project.

They keep saying about distributing inference, but we all know most of AI model are running on a closed source proprietary software owned by NVIDIA and they definitely don't support these kind of distributed inference thing on their library.
How are they handling big parameter LLM loading across distributed computing like these?

What are these project doing exactly? are they creating their own library or what? something that even AMD hardly could pull off within more than decades.

I hardly believe these kind of decentralized inference project, running an LLM on consumer grade hardware usually get you low parameter LLM that close to useless for anything.

[StreetWall Token]

This is an important question, and one that hits close to the core of what decentralization was always meant to address: power concentration.

You're absolutely right, we saw the same thing with banks, then exchanges, and now it's AI. Right now, the vast majority of large-scale AI models (like GPT, Claude, Gemini, etc.) are hosted, trained, and controlled by a handful of corporations. That means access, privacy, cost, and even the values that govern these systems are centralized by default.

The promise of decentralized AI, like what DecentralGPT is aiming for, flips that on its head. It’s not just about running models across distributed systems, it’s about democratizing who gets to train, access, and benefit from the technology.

If done right, decentralized AI could:

Lower costs by pooling global compute and removing gatekeepers

Increase privacy by letting users run or query models locally or peer-to-peer

Build trust via transparent training data and governance

Resist censorship in politically or commercially sensitive use cases

But it’s not without trade-offs. Latency, model quality, incentives for node operators, and protection against misuse are real challenges that still need solving.

Still, if decentralization changed how we bank and how we trade value, why not how we think?

Would love to hear if anyone here is actively building or exploring in this space. There’s massive potential if the right community rallies around it.

[coin-investor]

Lately I’ve been curious about how AI run across different systems instead of a single company’s servers. Realized we have DecentralGPT now and i find the idea quite interesting. It’s aimed at making AI cheaper, more accessible and more private.

But what do you think about this ? Will AI becoming decentralized actually make a difference ?

I like the idea; just imagine we will have minimal data breaches since the server will be distributed across multiple networks, rather than a single centralised server, and there are many more benefits for the AI community, like enhancement of privacy and resiliency.

This is very challenging to create, as the community must navigate numerous technical complexities to accomplish this. So far, we do not have this kind of AI, but the technology is evolving, and we may have this kind soon.

[gunhell16]

In this space, we’ve always known centralization as an obstacle, we’ve seen it in banking industries, exchanges and some other sectors. Now AI seem to be following the same path with a few companies controlling the computing power and access. Looking like those days before decentralization.

Lately I’ve been curious about how AI run across different systems instead of a single company’s servers. Realized we have DecentralGPT now and i find the idea quite interesting. It’s aimed at making AI cheaper, more accessible and more private.

But what do you think about this ? Will AI becoming decentralized actually make a difference ?

The question is, do all companies want that kind of thing regarding the AI revolution we're having right now? Because for sure, not everyone will want that kind of method,
and if there are some who adopt it, they might just be going along with the trend.

Also, I don't quite understand this decentralized GPT that the original poster is talking about. I'll probably have to first learn thoroughly what its pros
and cons are for the generation we have now.

[cabron]

Its still a machine so its always going  to need a human.

However if the AI is really that sophisticated like it can take care of itself that it will look for power sources, maintain itself and orders parts of fabricate something for its use that it need not humans, it might just be decentralized. Unless outside factors can destroy it like human. Nothing  still can stand on its own unless we allow it.

[bitgolden]

There will be some project sometime in the future where it is going to be just a project decentralizing AI power one day, but that's just not the one we are dealing with at the moment and it will be some other time. Which is why it's going to be something that doesn't exist today. If that happens, and there is a brand new huge one eventually, then it will be great for the long term.

However, what we have right now is not that great because the ones that already exist do not have that power or fame or attention. For this to work, something has to do a lot better before we can actually get a good result, and right now we do not have that kind of power on anything at the moment to get that type of attention or result at all.

[o48o]

If done right, decentralized AI could:

1. Lower costs by pooling global compute and removing gatekeepers
2. Increase privacy by letting users run or query models locally or peer-to-peer
3. Build trust via transparent training data and governance
4. Resist censorship in politically or commercially sensitive use cases
5. But it’s not without trade-offs. Latency, model quality, incentives for node operators, and protection against misuse are real challenges that still need solving.
6. Still, if decentralization changed how we bank and how we trade value, why not how we think?

Ok, i numbered your points to talk about them

1: Literally same amount of energy is being used for computing, so where are those costs reduced from? And how is that removing gatekeepers exactly?

2: How does it increases privacy if you need to trust some random people on the other side of that peer?

3: Elaborate please. Transparent training data is interesting idea, but that data is so huge that i am not even sure if it's manageable to govern.

4: So you are literally putting that censorship to the hands of other people. And do you think they want to be part of rendering something illegal?

5: I agree.

6: This is little vague. If you are talking about lack of hierarchy, for many people that's a foreign concept, even though we have those working models in nature already.

[Webetcoins]

Just like everything else in this space, it is a "let's see" kind of deal. I am not saying that it is going to be great, I am not saying it will be terrible.

I am saying, someone will eventually do it, and we are going to see if it is good or not.

Plus, it depends on who does it. A decentralized AI that is done by some dumwit will not be good at all, and nobody will love it.

A decentralized AI done by some revolutionary mind will be awesome and will be liked a lot for a while. Will it sustain and go long time?

We do not know. One thing we know is that AI uses a lot of energy, servers, GPU and many more.

If someone can tap into "everyone" market instead of just one, and use it packets in P2P form, they can unlock so much more power than even what openAI can afford.

[khiholangkang]

There will be some project sometime in the future where it is going to be just a project decentralizing AI power one day, but that's just not the one we are dealing with at the moment and it will be some other time. Which is why it's going to be something that doesn't exist today. If that happens, and there is a brand new huge one eventually, then it will be great for the long term.

However, what we have right now is not that great because the ones that already exist do not have that power or fame or attention. For this to work, something has to do a lot better before we can actually get a good result, and right now we do not have that kind of power on anything at the moment to get that type of attention or result at all.

AI also requires data centers for its research and must be constantly updated. It requires a lot of power to operate. Without power to manage it, AI will not function.

On the other hand, AI also needs to provide benefits to its contributors, just as Bitcoin fairly rewards miners. If such principles can be implemented for the successful development of decentralized AI, then we can safely say it is possible.

[Firstfrost]

There will be some project sometime in the future where it is going to be just a project decentralizing AI power one day, but that's just not the one we are dealing with at the moment and it will be some other time. Which is why it's going to be something that doesn't exist today. If that happens, and there is a brand new huge one eventually, then it will be great for the long term.

However, what we have right now is not that great because the ones that already exist do not have that power or fame or attention. For this to work, something has to do a lot better before we can actually get a good result, and right now we do not have that kind of power on anything at the moment to get that type of attention or result at all.

AI also requires data centers for its research and must be constantly updated. It requires a lot of power to operate. Without power to manage it, AI will not function.

On the other hand, AI also needs to provide benefits to its contributors, just as Bitcoin fairly rewards miners. If such principles can be implemented for the successful development of decentralized AI, then we can safely say it is possible.

Definitely! If AI doesn’t reward contributors then it people might actually lost interest, the recent one i’m aware of is DecentralGPT which rewarded its contributors with its token $DGC which got listed on Bitget. Also, i’m sure these AI projects would have made necessary requirements to function well but the issue which i think might arise is how well the ‘decentralization’ is accepted by people.

[khiholangkang]

There will be some project sometime in the future where it is going to be just a project decentralizing AI power one day, but that's just not the one we are dealing with at the moment and it will be some other time. Which is why it's going to be something that doesn't exist today. If that happens, and there is a brand new huge one eventually, then it will be great for the long term.

However, what we have right now is not that great because the ones that already exist do not have that power or fame or attention. For this to work, something has to do a lot better before we can actually get a good result, and right now we do not have that kind of power on anything at the moment to get that type of attention or result at all.

AI also requires data centers for its research and must be constantly updated. It requires a lot of power to operate. Without power to manage it, AI will not function.

On the other hand, AI also needs to provide benefits to its contributors, just as Bitcoin fairly rewards miners. If such principles can be implemented for the successful development of decentralized AI, then we can safely say it is possible.

Definitely! If AI doesn’t reward contributors then it people might actually lost interest, the recent one i’m aware of is DecentralGPT which rewarded its contributors with its token $DGC which got listed on Bitget. Also, i’m sure these AI projects would have made necessary requirements to function well but the issue which i think might arise is how well the ‘decentralization’ is accepted by people.

The big question is, how many users will this AI have, and how much income will contributors receive in return? Of course, it must be calculated to be profitable in order to maintain decentralization, because the cost of power requires a considerable amount of expenditure, although it depends on how much power they consume.
What if there are few users? And I noticed that their tokens also amount to 1T with a very low value, which is enough to make many people skeptical.

[Odohu]

How do you think there could be decentralization in AI, something that is patented? I don't know which resources you are getting your information from but AI cannot be decentralized and all the AI service we have seen have people who own them. What we will see if they want to lie to us about decentralized AI service, is companies pretending to offer such but in reality they will be heavily centralized.