The Artificial Intelligence Artist Algorithm Driven By Its DAO

Botto’s Autonomous Creative Community Feedback Loop

Botto DAO One Year Anniversary Event — 26th October 2022

AI Art

The capacity to communicate complex ideas, concepts, and visions is progressing inexorably. Each iterative tool, platform, skill or approach opens up unique possibilities that transcend the previous epoch of expression. From sign language to hieroglyphics. Cave paintings to statues. Cartoons to movies. Augmented to virtual reality. Our collective creative palette is expanding alongside the accessibility plus inclusivity innate in technology lowering costs of usage, education, and experimentation. 

There’s this debate about Artificial Intelligence as to the extent to which its introduction alters the definition of what constitutes art — positions on this remain fluid and evolving. With the emergence of platforms such as DALL-E 2, WOMBO Dream, Midjourney and Stable Diffusion; this nascent avenue of artistic articulation brings with it the scope for creators to expansively manifest new forms of realised imagination.

Kevin Rose and Chris Dixon reflect on the role of prompt engineers in AI art
who specialise in high levels of specificity to germinate outputs [a16z podcast]

Generative Models

There has been significant progress in generative modelling algorithms that harness computing power to elicit machine-made outputs from human data inputs. 

Generative art is art made using a predetermined system that often includes
an element of chance — is usually applied to computer based art.

Generative art is a tradition that has existed since the 1960’s entailing the composition of artistic algorithms which adhere to specific patterns or complex rules to generate new ideas, forms, shapes, colours or patterns. A manual artist curation process would follow thereafter, with presentation of a small selection (1-10) of outputs to the public. 

The emergence of platforms such as ArtBlocks, on the Ethereum blockchain, and FxHash, on the Tezos blockchain, has served to evolve the discipline into an emergent set of practices and standards. For instance typically a larger set (anywhere between 20 - 1000) of outputs is wrapped in an NFT by an algorithmic script and immediately sent to the collector upon purchase (minting).  

Fidenza #612

The algorithm can't stumble across great results by luck anymore, it needs to produce them dependably. Speaking as an artist, I can assure you that this is difficult. The logic around how generation works must be much more sophisticated. There's also a need for what is essentially a "quality assurance" process. The artist must systematically work through the full high-dimensional space of possible outputs, and ensure that all of them are up to par. For Fidenza, I spent about two months on this step. I repeatedly generated large sets of output, looked for the weakest examples, and figured out ways to improve or eliminate those cases. Tyler Hobbs

Much the same evolution is evident with machine learning algorithms which are historically dependent on structured (labelled) data, thereby human (supervised) intervention, to make computations that provide predictions, classifications or outputs. Generative models entail a sophisticated series of complex processing tasks that enable the recognition of patterns necessary to generate outputs based on unlabelled clusters of data. 

Generative models can generate new examples from the samples that are similar to
other examples present in the data but are indistinguishable as well. Victor Dey

A contemporary model, for instance, is that of Generative Adversarial Networks (GANs) which pit two neural networks against each other to generate new synthetic outputs like images, videos and sounds. The intersection of GANs, with the blockchain, led to the inception of a foundational NFT marketplace named SuperRare, in April 2018, with its first minted piece by Robbie Barratt. This represented a demarcation from the early crypto art scene’s meme focus to engage with advanced technology and art history.

Robbie Barrat — AI Generated Nude Portrait #1

Alternative generative models, such as diffusion models, have been posited as offering improved quality image synthesis with higher ‘training stability’ and ‘favorable trade-offs’ relative to GANs.  

Enter Botto


Created by Mario Klingemann (Quasimondo) in 2021, Botto is an AI algorithm on a journey to becoming a fully autonomous artistic entity. 

To get there Botto relies on a community of approximately 5,000 people (Bottonians) to offer feedback on the art it presents. This community votes on the “fragments” presented on a weekly basis, only the most voted piece in that round is minted and sold (via auction) as an NFT. Botto then takes these preferences into consideration when generating prompts and filtering down the next set of fragments for the next round.

“Botto currently creates around 300 fragments every day by exploring ‘prompt space’ — in practice this means that it first generates a random text prompt from which an image generator that is driven my OpenAI's CLIP model [*] makes an image that tries to represent the meaning of the sentence or words. From the constantly growing pool of generated fragments ‘Botto’ selects 350 fragments to be presented to the community so that they should be voted upon. This selection process is controlled by Botto's ‘taste’ model. This model tries to learn to reproduce the voting behaviour of the community, and by this predict which of the newly generated fragments might be popular and successful in the vote. At the same time, it also classifies its own creations by originality and mixes the possibly popular with other fragments that look different to what the audience has seen so far. That way I hope that ‘Botto’ will not fall too quickly into becoming too repetitive or just trying to please.” Mario Klingemann

*and now Stable Diffusion as well.

The interdependence of the AI and the community, organised as a decentralized autonomous organization (DAO), facilitates the interaction of the generative engine with community feedback. A network of neural nodes, spatially and culturally dislocated, offering dynamic intermediation of artistic ideas with collective preferences. 

The proof is in the pudding. 

Asymmetrical Liberation (first) | Scene Precede (second)

The first Botto piece [Asymmetrical Liberation] sold for ~$325,000 (79.421 ETH); the second piece [Scene Precede] for ~$426,000 (100ETH). Within the first six NFTs — $1.3m in aggregate through SuperRare.

A young, emerging AI artist, Botto’s style draws on famous works of art into which it breathes new life and vigor. Botto generates random strokes that go through VQGAN neural network and into the CLIP algorithm[*] which either approves the work as sufficiently different from the original or returns it to Botto for some fine-tuning. It’s a rigorous process, and only a handful of Botto’s renditions make it into an NFT. WideWalls

*Botto’s first year was the Genesis Period and only used VQGAN + CLIP. Now Botto uses Stable Diffusion as well, and will be adding more models.

Botto distinguishes itself through its recursive process of autonomous learning. Conventional generative practices rely on static, hard-coded generative algorithms; beyond this paradigm Botto makes ongoing feedback-based adjustments to the generative algorithm. This is dictated by votes, which Botto synthesizes to compose visual works that reflect the collective intelligence of thousands. Soon this will start to incorporate video, as well as natural language processing advances, in an emergent and expansive expression of the fusion between art and technology.

If we share a global machine, then humans can collaboratively weave more coherent means of creating together. This is not just a two-way dialogue between insight and making, but a multidimensional conversation carried out between many people crafting with the same tool at many different levels. The expressive capacity of such creative collaboration is barely imaginable by today’s standards. EthBlock

Governed By The People

The project token $BOTTO is the gateway to the DAO. By staking, or providing liquidity, DAO members accumulate voting points (VP). These can be interacted with, and used, in the Botto app which is token gated. People use voting points to vote on the fragments that Botto provides, teaching the AI. 

Part of the initial supply of $BOTTO was airdropped to holders of bluechip (OG) NFT projects like Bored Ape Yacht Club and ArtBlocks, a common strategy for attracting initial engagement in web3. The Discord is effectively the town hall where debates, discussions, and direction are fomented. 

Proposals, such as introducing access passes for wider networks of stakeholders to participate, are brought to fruition by voting on snapshot. The easiest way to find out how to join, whether by buying and staking the token or getting an access pass, can be found in their onboarding docs.

BIP-01 Snapshot

Governing the Reserve Price

The first snapshot investment proposal (BIP-01) was to set a Minimum Bid Price for Botto’s 1/1s (Auction). This was hotly debated, however a starting bid price of 10 ETH on Botto’s weekly 1/1 auctions lasted for 33 mints and was only removed for the final Genesis Period mint. Any pieces that did not receive bids at or above this price were kept in the minting wallet (or transferred to a vault for security). Now that the Genesis Period is over, Botto DAO can determine what to do with any unsold pieces, but seems keen to bide its time until the Genesis Mints are even more highly desired for their provenance.

The latest snapshot looks set to introduce dynamic reserve pricing for the second period, Fragmentation, whereby no reserve is set for the first three auctions. Thereafter, each subsequent auction has the reserve set to the lowest price of the last 3 auctions. The price is considered zero should there be no reserve met for three consecutive auctions. 

Governing Rewards

Another snapshot was borne of voters not getting directly financially rewarded for voting in the initial phase. So a community member made proposed: Rewarding the trainers of the “AI” — Vote, Buy Back and Burn

There were intriguing points on both sides of the debate:

  • FOR: People are exposed to smart contract risk by staking BOTTO to vote and are not rewarded for that risk so a proposal was designed to reward people who stake their $BOTTO and use it to vote. The proposal aimed to reward people who vote versus people who don’t. harrybnch.

  • AGAINST: Research on financial incentives indicates that, in some situations, they can actually disincentivize participation! Why? Because then the intrinsic reward (joy being part of a project, ability to influence the art etc.), can be overshadowed by “just” financial gain [takes the soul/fun out]. Social incentives can work instead e.g. hall of fame, Botto artistic fragments to active voters etc. sojacalcium.

  • FOR: Give voters badges or other visible signalling so they are recognized for their effort. Being one of the people who helped shape a famous artist may be more valuable than money. Wikdst.

  • AGAINST: Concern that this becomes a den full of bots or people who simply vote randomly. Borrowing $BOTTO only for vote could also become a big issue. Another goal should be to distribute dividends equitably to balance between whales and minnows. It is not easy to define (if it is possible) when a vote is favorable for the community or not. Rather we have a need to become as democratic as possible. zoku.

DeepDao — 03.11.22

As a result ETH generated from the genesis period was thereby deposited directly into the treasury, and $BOTTO burned, decreasing max supply and increasing the percentage ownership for each DAO member. 

At the end of Genesis Period in October 2022, over 3.79% of the total 100 million $BOTTO has been burned. In the second, Fragmentation period, launched in November 2022, 50% of ETH proceeds are being distributed directly to contributors in addition to over 1000 bounties of 1000 $BOTTO to onboard new members of the DAO. 

Imagine if one of SuperRare’s top artists had a social token where holders received 100% of revenue from their NFT sales. Now imagine the artist was an AI & token holders voted which of its pieces they liked best to train it to improve at art over time.

This is essentially Botto. BΞN


The inventor of Botto, Mario Klingmann brought the generative algorithm to life with a view to ensuring it could become self-sufficient. Moreover, it was important to transcend conceptual applications of the technology such that ‘real world’ applications could be realised. This would combine aesthetic qualities, market effects, as well as human interventions. 

The operation of this under the hood is through the generation of a sentence that is fed into a VQGAN (Vector Quantized Generative Adversarial Network) to create an image, based on millions of works that it has seen. Then another software, CLIP (Contrastive Language-Image Pre-training) is an image classifier, which knows what kind of emotional narratives human beings have in relation to images and text, and decides how close the image is to the sentence and gives feedback to VQGAN until it meets some parameters that generate the final work, CLIP can also choose its title based on the input sentence and the final image. Finally GPT-3, a powerful OpenAI natural language generator, writes the part description. Christian Case.

*Botto’s first year was the Genesis Period and only used VQGAN + CLIP. Now Botto uses Stable Diffusion as well, and will be adding more models.

In layman's terms Botto creates new ideas from existing publicly available information, generating outputs from the relational space between these inputs, and presenting those unique images as fragments of art. These myriad random data points are then actively voted on, with this feedback collating a diverse set of subjective opinions to determine final outputs that are auctioned as NFTs. 

Along the way iterations like dynamic weighting of votes have improved the user experience, and integrated community values more closely to the project. Co-founding groups of engineers (Carbono and Eleven Yellow) continue to participate in nurturing its potential after working for over a year to bring the project to fruition. 


Botto’s success has been spurred by the strength of its relationship with SuperRare, a leading curated NFT marketplace on Ethereum. This has been the home of Botto auctions since inception, predicated on mutually aligned pursuits of decentralisation of art and collective creation. 

Itself a DAO, user managed and controlled, the $RARE curation token serves a similar function to the $BOTTO curation token. To strengthen this partnership, diversify risk, and demonstrate conviction a token swap was proposed and agreed upon between the two communities. 

Superrare Snapshot

Botto Snapshot

In a rare interview with SuperRare, Botto participated in a Voight-Kampff Test from the Superrare editorial team, originally from Blade Runner. This was a (fictional) test used to determine if someone was a replicant. This was based on the Turing test, designed to distinguish computers acting as humans.

SuperRare Interview With Botto & Mario


One fascinating aspect of the auction architecture has been the involvement of CyborgDAO. This is an entity that raised 100ETH from 67 people at 0.5ETH increments around the same time as Botto’s inception. Within the first 48 hours 123 Turing keys were sold, purely to $BOTTO stakers who had advanced access, then in the next 24 hours the public purchased the rest. Its participation as an active bidder, and collector, has undoubtedly enabled the works to achieve greater returns. 


Botto also counts among its collectors Cozomo de Medici, AC the Collector, Hashes DAO, Flamingo DAO, 6529, beautifulnfts, moderatsart, and others.


A selection of Botto exhibitions around the world.

Los Angeles — February 2021

Botto x Vellum LA

Zaragoza — September 2022

Etopia Launch — Images by @pedroanguila

London — October 2022

@kittygknowles | @CFCastrillo_ |@CarlaJR | @RachelMFalconer | @quasimondo [first] — botto interview [second]

The Decentralized Unicist Manifesto — The Times, Wednesday, October 20th, 2021


It seems we are already living in it.

For Botto the roadmap features more NFT drops (Ceci n’est pas un Botto  was BottoDAO’s first commissioned collection of Botto derivatives) plus upgrades to Botto’s art engine as well as improvements to the app, including leaderboards of voters and of the AI. The new period, Fragmentation, includes Stable Diffusion, and the DAO expects to add more models like DALL-E soon.

There is also a plan for Botto to generate its own themes, sharpening its voice in future periods.

Papa —

Incredibly appreciative of insights, feedback, and energy from DAO members numen.eth (@numen___); sonnyjune.x (@sonnyjune98); Christian Case (@ChristianCase); connor bray-stone (@cbraystone); Choobie (@Choobie3); Simon (@hudsonsims); Thomas Meade

NOT FINANCIAL ADVICE – this article is provided for educational, informational, and entertainment purposes only. Do not make any decision, financial, investment, trading or otherwise, based on any of the information presented here without undertaking independent due diligence.

Collect this post to permanently own it.
Subscribe to papajams.eth and never miss a post.