Artificial Intelligence Booming Year Is Behind Us: What's Next in the AI Industry?

Artificial intelligence-based tools for image, text, sound and content creation in general marked the year 2022.

The year 2023 brings a handful of opportunities, but also challenges. What artificial intelligence is expected to bring us by the end of 2023, find out below!

More artificial intelligence apps, more problems

After the success of Prisma Labs’ selfie app Lensa, we can probably expect a lot of similar apps.

At least some of them – if not all – will be able to be tricked into creating images that are not for showing colleagues at work, and excessive sexualization and changing the appearance of women will continue to be a problem.

The expected integration of generative artificial intelligence into consumer electronics will only amplify the current good and bad aspects.

However, 2023 should be the year during which generative artificial intelligence will finally shine. But in order for that to be possible, it will have to either start bringing in a lot of money, or have a meaningful impact on the everyday life of the public.

It is to be expected that there will be attempts in both directions, with variable success.

Artists will seek exemptions

DeviantArt has released an AI and Stable Diffusion based art generator, customized for its community’s artwork. The move was met with loud opposition from long-time users of the service, who criticized it for the lack of transparency surrounding the use of works of art to train artificial intelligence systems.

The creators of the most popular systems – OpenAI and Stability AI – say they have taken steps to limit the amount of harmful content their systems produce.

But, judging by what appears on social networks, it seems that there is still a lot of work to be done in that segment, which will require the participation of the people whose work is being used.

Stability AI has already bowed to public pressure and indicated that it will allow artists to be denied permission to use their work when training the next generation of Stable Diffusion.

Those who wish to do so will be able to do so in a few weeks at HaveIBeenTrained.com.

OpenAI does not currently offer such an option, preferring to partner with organizations such as Shutterstock to secure licenses to use parts of their catalogs.

But, it wouldn’t be strange if they soon followed the same path as Stability AI.

It is possible that, in the end, they will be forced to do so by court decisions.

In the United States, a joint lawsuit has already been filed against Microsoft, GitHub and OpenAI for copyright infringement by allowing Copilot, GitHub’s service that suggests which lines of computer code should be written, to use parts of licensed code without disclosing who wrote it.

At GitHub – perhaps precisely because they anticipated legal problems – they recently added settings that prevent the display of publicly available computer code in Copilot’s suggestions.

They also plan to introduce a feature that will show where the suggestion came from.

However, both of these measures currently do not guarantee that the failure will not occur.

Open source and decentralization will continue to grow

During the year at the end, only a handful of companies dominated the artificial intelligence stage, led by OpenAI and Stability AI.

However, during the next year, more new challenges could appear because the creation of new systems could become available outside specialized laboratories rich in resources.

If such a development occurs, it is possible that we will witness stricter supervision of the system being built and implemented.

An example of such efforts are, among others, the large language models produced by EleutherAI and Big Science, which is behind the startup Hugging Face.

Stability AI has launched several such communities, such as Harmonai (for music generation) and OpenBioML (biotechnological experiments).

Money and know-how are still needed to train and run sophisticated AI models.

ChatGPT’s chatbot is estimated to cost around US$3 million a day to operate.

But decentralized computing could throw down the gauntlet in the face of traditional data centers as open source computing projects mature.

BigScience has taken a step in this direction with the recently launched Petals project, where people can lend their computing power (similar to Folding@home) to run large language models that would otherwise require an advanced graphics card or web server.

Large laboratories will continue to have an advantage as long as the methods and data remain their property.

Pressure from regulatory bodies will increase

Regulation such as the European Union’s Artificial Intelligence Act could change the way companies develop and implement artificial intelligence systems in the future.

Regulation and increased supervision are necessary because generative artificial intelligence is still prone to errors, so it is not advisable to apply it in areas where it could cost a lot, such as healthcare.

However, artificial intelligence systems are already making decisions with serious moral and ethical implications.

Arguments over the rules and litigation should be expected before the fines and indictments begin.

Artificial intelligence systems are currently divided into four risk categories, which differ in the degree of supervision and requirements they must meet.

High-risk systems (algorithms for robotic operations and determining credit rerating, for example) must be at a certain legal, ethical and technical level before they can enter the European market.

On the other hand, for low-risk systems (spam filters, computer games with artificial intelligence), it is enough to inform users that they are dealing with artificial intelligence.

It is not impossible that there will be companies that will try to push their services and platforms into that category in order to reduce costs, even though they belong to one of the riskier ones.

Will there be investments?