The Radical Accessibility of AI
GeMs are making it possible for everyone to be an AI-augmented producer
There are many ethical and societal questions regarding the use of AI systems. For instance, how will the continued doubling of computer energy use affect the climate? Or, how might we mitigate the negative impacts of AI on the economic livelihoods of people?
In this post, I consider AI systems in terms of equity: ie, whether the power of AI systems is equitably distributed among all people. To use AI systems, people need to find them valuable, enjoyable, affordable and easy to use. The last factor has seen enormous success in the past year: several extremely powerful AI systems now have user interfaces that are as easy to access as popular consumer applications, such as Facebook and Twitter. In other words, AI is now Radically Accessible.
This level of radical accessibility stands in marked contrast with the use of AI systems in past years. Just last year, using generative models (like GANs) required large amounts of data and significant technical skill. Recent tools, like GPT-3 and DALL-E by OpenAI, have made powerful AI models available to the public through easy-to-use interfaces that are relatively low-cost. Even free!
In that spirit, why don’t you generate an image now, if you haven’t? All of these services are available for use, with some degree of free use:
https://beta.dreamstudio.ai/
https://openai.com/dall-e-2/
https://www.midjourney.com/home/
Now, another aspect of accessibility is the transparency of the system. Here, OpenAI has been surpassed by the StabilityAI image generator, which is open source and enables people to browse the actual data used for training.
Yet, the radical accessibility of AI systems is raising some concerns. For example, if everyone has access to powerful AI image generators, will we see an increase in the number of deepfakes? If everyone can access large language models, will we see an increase in the spread of misinformation or spam?
It is important to understand the upsides and downsides of radical accessibility in AI systems. But it is also important to understand that the cat is out of the bag. The dramatic success of open-source models makes it seem especially unlikely that legal challenges will be able to limit the availability of AI systems.
If the power of AI will be with the people, how can we ensure that this power will produce the most good? This question is not easily answered. But one example in our community aims to extend the power of AI image generators to unlikely early users: for instance, traditional craftspeople in India. How might image generators be used by craftspeople to develop new designs — e.g. for jewelry, baskets, furniture, architecture, etc?
Another example might be the use of AI writing programs helping people with special needs communicate more easily. In any case, we shouldn’t be surprised to see powerful AI systems show up in surprisingly places.
What other opportunities or risks so you see that might emerge from the radical accessibility of powerful AI systems?
This is indeed a highly exciting area for me. Having worked with artisans and crafts people since the early 2000s, I find it fascinating that finally we may now have the "radical accessibility" to design that these people were denied all these years. We always talk about technology being a killer of the hand-made ecosystem but never did we think that an advanced technology like AI could actually become the enabler. I look forward to working in this space!
And yeah, thanks for including some of the generative designs that I worked on here. Cheers!