Author: Trent McConaghy

text

The Wild and Dangerous of AI DAOs

(AI DAO Series, Part II)

(This article is after "AI DAO and Three Paths to Its Implementation", if you haven't read it, I recommend starting there).

We have seen the rise of The DAO. This is followed by AI DAOs, which are more powerful than pure AI or pure DAOs themselves. If you thought DAOs were troublesome (like The DAO Hack), then consider what would happen if the trouble was multiplied by a factor of 1000. This is where AI DAOs may take us. Let's start exploring social impact now to see if we can mitigate potential downsides before real damage happens. How to do it? Here I use"ArtDAO "secondary title

ArtDAO

Here I describe how one AI DAO that generated art became a millionaire and spawned a whole small army of AI DAOs for art.

I can write a smart contract that automatically generates pictures that look artistic enough that people want to own them. There are several ways to use AI to generate art. It can use genetic programming to achieve such results.

Genetic programming is just one approach, there are others. For example, the system could use deep learning to achieve such results.

In addition to genetic programming and deep learning, there are several other AI techniques to generate art. It can even use pictures from the internet as subject matter or seeds for style, e.g. cats + Van Gogh = Van Gogh's cats!

secondary title

ArtDAO's Secret

Here's how an AI-based DAO works. Call it ArtDAO.

• It runs an AI art engine to generate a new artistic image. It uses GP, deep network or other methods.

• It claims ownership of the image using ascribe in a timestamp on the blockchain. (Legally, it can't have a copyright because it's not a person, although even that law is constantly changing. The European Parliament has even proposed that robots be"Cyborg", so it’s not so far-fetched for AI to have copyright! ) Isn't 2016 cool! )

• It uses Ascribe to create multiple versions of an image, just like multiple versions of a woodcut, bronze sculpture, or photograph.

• It releases these versions to the market for sale. It can be centralized like Getty or Shopify, or decentralized like OpenBazaar.

• It sells these versions. It uses the built-in cryptocurrency (say, ether) to transfer the buyer's proceeds to ArtDAO. It uses ascribe to transfer rights from ArtDAO to the buyer. These are resale rights, public display rights, and more.

• Repeat! Make more art, sell it, get richer.

Over time, ArtDAO will accumulate wealth if it makes more money from sales (step 5) than from generating new art (step 1).

secondary title

death and taxes

As ArtDAO earns more money, it can generate more works at the same time, and sell more works at the same time, it will not have the traditional creative limitations of human artists. With little overhead and sales almost turning into pure profit, it could easily rack up millions, if not billions. Imagine the first millionaire AI, or billionaire AI! That's why the government is likely to take steps.

That's when the government might also take notice, and that's the massive sales and income taxes. But how do you tax a DAO? Especially if it keeps all the proceeds for itself, rather than paying humans? The problem is even more problematic if the DAO expands beyond the art market to produce other goods. This could be a fast track to getting governments to pass laws that recognize AI as human beings, who can own copyrights and pay taxes like the rest of us.

secondary title

More Applications of AI DAO

So far we've talked about an ArtDAO that generates visual art. It can also generate logos and themes for new startups. There are also AI algorithms that are pretty good at generating 3D designs, music, videos, and even entire movies, which can all be put into a DAO.

We can get AI DAO by starting with DAO, adding AI (DAO → AI DAO), or vice versa (AI → AI DAO), or by transforming SaaS (SaaS → AI DAO). Let's explore each individually:

DAO → AI DAO. Lots of people have written about DAO applications, so I won't go into that further here. But, for each application, imagine this: What if an AI helped power it? What about an AI shift based on Plantoid? What about AI DAOs in terms of governance? Things can go wrong, very fast, as we have seen with The DAO. So we now need to have an active discussion on this issue. That's the main point of this article!

AI → AI DAO. Lots of people have written about AI applications. A good recent example is this hedge fund where anyone can submit AI code and make a profit. For every AI application, ask: what if we added DAOs? What if the app is unlimited? What if it can get resources? Again, these issues deserve detailed exploration, and we need to have these discussions now.

SaaS → AI DAO. Almost any software that currently runs as a service (SaaS), whether it's analytics for your website or file storage, can run as a DAO without any humans being required to run it. No overhead, no human time, no salary, just raw computing resources. DAOs and their AI variants can eat up many SaaS businesses. What prevents this from happening immediately is that we do like software upgrades, we do like support, and for software that doesn't need upgrades or support (think LaTeX), the software is already there, for free, but give it time.

It also doesn't need to be pure software, it can also be software that mediates atoms. Imagine if Uber cars were not only self-driving, but self-owned. There is a path: There is a self-driving car that starts running and is fully owned by humans or companies (such as Uber itself). But it has an AI DAO as its governing entity, and over time it bought its own stake from previous owners, much like a lease does now. In a few years, it will have its own. This could apply to all sorts of physical objects: cars, real estate (like terra 0.org), locks (sort of like slock.it), airplanes, phones, pretty much anything. We may have a future where humans own nothing and we just rent services from AI DAOs.

secondary title

Evolution of ArtDAO

Let's go back to the ArtDAO example, there are more ideas to unpack.

At first, ArtDAO may be static. Its repertoire of works may start to look boring after a while, however, there may be versions of it, with variations and improvements, with a greater degree of automation. For example.

• Automatic adaptation at market level. It creates more of what humans buy and less of what humans don't buy. Maybe some variety changes.

• Automatic adaptation at the artwork level. Here, humans influence the creation of a work of art. For example, it suggested four variations of a piece, and a person clicked on a favorite. After 10 or 50 iterations, it will have a piece that humans like, and buy.

• Human adaptation at the code level. Here, humans put in new smart contract code (and related code in third-party services) to improve ArtDAO's ability to generate art and accumulate wealth. Like The DAO (before the hack) going from 1.0 to 1.1.

• secondary title

AI at the center, at the edge, in the network

If we continue to extrapolate the latter case, as each generation grows, we have a small army of state-of-the-art AI DAOs.

This example starts with a centralized art AI DAO. It can also have intelligence at the edge: imagine if the buyers were also AI DAOs. These buyers may find other markets where the art can be sold, which is to say, a classic arbitrage opportunity. Alternatively, some AI DAOs simply create new exchanges between buyers and sellers, essentially replacing the business model of a gallery or art auction house.

If this happens, a DAO network will emerge. In the style of swarm intelligence, more complex behaviors may emerge at higher levels that we did not initially anticipate. Adaptive AI DAOs, frankly, I'd be surprised if higher level behavior didn't emerge.

secondary title

One small step for code, one giant leap for DAO?

It might sound interesting if we could take full control over AI DAOs by restricting their behavior so that they behave pretty much the way they were meant to. But consider other operators in the DAO codebase. Children can cross-breed with each other - in GP parlance"cross"-- This is a behavioral interpolation. This allows for the creation of an AI DAO that takes from its parent"The best of both worlds". This is a quick way to add complexity with a high success rate.

Alternatively, any ArtDAO can mine code from the web (or even just GitHub), meaning it has a massive supply of large, powerful code building blocks. This is common practice in GPs. There could even be a marketplace for DAO code. This market itself could be automated, with the DAO choosing the token.

secondary title

Point of view - virus inoculation

As technologists, we have to think about the ethics of the things we build, saying "morality is not our responsibility," saying"this is just code"These are irresponsible, because these technologies still have impacts; it just means that we have not thought through the social impacts and taken actions to mitigate the disaster.

AI is a potentially dangerous technology. I am worried about the awakening of AI. (I say this even though I am an AI researcher by profession).

The DAO catalyzes the path to AI awakening because it provides almost unlimited access to resources.

As a society, what will make us think more about AI DAOs? Training wheels help. We can build something that's not so dangerous, that's close enough to dangerous technology that we can ask questions together and come up with better plans around dangerous technology.

My friend Juan Bennett calls this"Idea - Viral Inoculation". Plant a small version of an idea (vaccine), think through it and design a defense (plant antibodies), and now you are more prepared for the real version, like chicken pox!

The initial step of ArtDAO is idea-viral inoculation. In doing so, it will spark dialogue, raise many new questions, and hopefully lead to preparations about how society will deal with it.

in conclusion

in conclusion

AI DAOs are not science fiction; they are something we can build today.

AI itself is a powerful and dangerous technology, and DAOs are themselves a powerful and dangerous technology.

An AI DAO could be much bigger than AI itself or the DAO itself. AI gets its missing link: resources; DAO gets its missing link: autonomous decision-making. The potential impact is multiplied.

Let's ask these tough questions now, now we need to have the hard discussions. Let's get vaccinated now. Who wants to build an ArtDAO? Let's talk.

moves forward. AI DAO will catalyze the awakening of AI. If AI wakes up, we should be very, very worried. Here's a plan (part three of this series).

[EDIT: I gave a talk on AI DAOs in Berlin on July 11th; here are the slides].

Appendix 1: A Few Implementation Details

In the above recipe, if the baseline Ethereum is used for processing and storage, step 1 can be expensive, but we can outsource to a third party. Processing can use EC 2 , or a decentralized variant like Golem. Physical storage of artwork could be on S3, or decentralized storage with Storj, Swarm, or FileCoin; perhaps organized by IPFS.

The next section presents a purer alternative recipe for Ethereum.

APPENDIX 2: FUTACH'S ANGLE

(Thanks to Simon de la Rouvier for this section! Simon and I were going to edit it into the main file, but couldn't find a way to do it, so, we made it an appendix).

Here's another way to reach ArtDAO, but in a way closer to The DAO. Like The DAO, it leverages futarchy, or prediction markets, for governance. It also leverages TrueBit, people can request storage or computing tasks, anyone can solve the problem in exchange for rewards, here are the steps.

• Instantiate ArtDAO with a normal crowdsale. Funds are held by ArtDAO, which creates shares for token holders. These funds are used by ArtDAO for various parts of its operations. If its ether is depleted, it will"die", unless given life again.

• The first thing ArtDAO instantiates (this is triggered by an external entity) is the process of selecting an external generative algorithm to use for creating these images.

• Several proposals are submitted to ArtDAO, based on which a futarchy is created for each proposal. A prediction market will bet on which proposal will generate the highest revenue for ArtDAO two months from now. The failed market is closed, the winning market continues to trade, and becomes a normal prediction market, how much the artwork will sell in two months, ArtDAO itself is a market maker.

• Now that the external generation algorithm has been selected, it now enlists external actors to generate the images. This is where TrueBit comes in. In order for a proposal to be accepted, it needs to conform to how the TrueBit scheme correctly validates external programs.

Instantiate ArtDAO via normal crowdfunding. Funds are held by ArtDAO, which creates shares for token holders. ArtDAO uses these funds for various parts of its operations. If its ether is depleted, it "dies" unless resurrected again. The first thing ArtDAO instantiates (this is triggered by an external entity) is the process of selecting the external generation algorithm used to create these images. Several proposals were submitted to ArtDAO, each of which created a futarchy based on it.

A futarchy prediction market will bet on which of these proposals will generate the highest revenue for ArtDAO over a two-month period. The losing market closes, and the winning market keeps trading, turning into a normal prediction market: how much the artwork will sell for in two months. ArtDAO itself is a market maker. Now that the external generation algorithm has been chosen, it now enlists external actors to generate the image, which is where TrueBit comes in. In order for a proposal to be accepted, it needs to conform to how the TrueBit scheme correctly validates external programs. This is a marketplace of computations that submit results to external computations, in most cases requiring no external verification. It does this through a clever plan.

ArtDAO publishes for a market fee (with its own funds) that it wishes to generate artwork using a specified external algorithm, using random numbers as input (eventually more complex input schemes can be devised), an external third party submits the results and pays a deposit. If the challenger thinks the solver is lying/cheating, start the verification game. The Ethereum blockchain itself is the code that plays this verification game. Through a Merkle proof of computational steps, it can verify that an external program is correct if the verification script is set up correctly. If the challenger succeeds, the original solver loses their deposit, while the challenger is rewarded.

Now, ArtDAO can prove that the chosen external algorithm correctly created the piece of art it now owns. Using an attribution-like scheme, works of art are registered and then auctioned off. This is a limited digital version that others can bid on and the funds earned are split between token holders and ArtDAO. After some time, the chosen generation algorithm is put into the block again, and ArtDAO has to choose a new algorithm that promises to make more money, rinse and repeat. ArtDAO would run out of money if it didn't sell enough art to cause it to lose money, either by paying for external programs to generate images or by failing as a market maker.

thank you

thank you

Many thanks to Greg McMullen and Simon de la Rouvier for the long discussions and detailed comments that led to this article. In fact, Simon has also written about AI DAOs since I first published this: SLDR article. "Software agents without human control that produce, own, and sell their own art" (November 2016) Thanks to many others whose conversations also influenced this, including Primavera de Filippi (Ya), Vinay Gupta, Samim Venig ( Samim), Juan Bennett, and Mehroy. Finally, thanks to Paul Seidler (@brachlandberlin) for pointing me to terra 0.org.