Unresolved AI Issues – When Will They Be Resolved, No One Knows

As the soap opera surrounding Sam Altman, the CEO of OpenAI, unfolds in recent days, with him leaving and then returning to the company behind the increasingly popular ChatGPT, while everyone praises this new generative artificial intelligence, few are currently addressing its issues, and even fewer will say that it is not actually true artificial intelligence.

Generative artificial intelligence is a type of AI that can generate new content, such as images, videos, music, or text. It does this by learning patterns and relationships within the data it is trained on, and then using that knowledge to create new content that resembles the original data. It is not intelligent and does not know anything that has not been stored in it; it knows only what it has been trained on or fed.

Some argue that true intelligence requires the ability to understand and apply concepts, logical reasoning, and decision-making based on context and intent, which generative AI systems may never be able to achieve. Others argue that generative AI systems exhibit a form of intelligence that is unique and valuable in its own right, and this is where battles rage among so-called experts. Regardless of whether it is real or not, this type of artificial intelligence is currently proving to be a good technology or tool in areas such as art, design, and entertainment.

As mentioned, generative artificial intelligence is trained by training human content based on massive datasets, and since it is not perfect, significant problems arise during training, including privacy, regulatory compliance, copyright, transparency, and bias.

Generative AI Laws and Frameworks

Although no major generative ethical framework or AI policy has been adopted into law at this time, several laws are in the works.

The European Union has made the most progress in regulating generative artificial intelligence, with Italy even briefly banning OpenAI until the company improves its data privacy standards. The EU’s Artificial Intelligence Act is a proposed law that would categorize AI applications into unacceptable risk, high risk, and low-risk categories, with particular attention to generative artificial intelligence and concerns regarding copyright/ownership.

While the United States does not have an official AI law in the works, several frameworks and best practices have been established that indicate legislation could come into effect in the future. Examples include the Biden administration’s draft for an AI Bill of Rights, the NIST framework for AI risk management, and guidelines for copyright registration for AI-generated content.

The United Kingdom is likely to continue regulating artificial intelligence at a slower pace than the EU but faster than the United States. The country already has a policy document titled Regulation of Artificial Intelligence: A Pro-Innovation Approach, which summarizes its plans for regulating artificial intelligence.

Generative artificial intelligence is new and unregulated, meaning there are many ways it can be misused. Here are some of the biggest ethical issues surrounding generative artificial intelligence today.

Copyright Issues and Stolen Data

For generative AI models to regularly produce logical human-like content, these tools must be trained on vast datasets from various sources. Unfortunately, most AI companies have concealed this training process, and several have used original artworks, content, and personal data from creators and other consumers in training datasets without the creators’ permission.

MidJourney and Stability AI Stable Diffusion are two tools that have come under fire for these issues. Personal and corporate data of other types have also been inadvertently introduced into generative AI training algorithms, exposing users and corporations to potential theft, data loss, and privacy violations.

Ethical data use becomes paramount when considering that several generative AI tools appropriate copyrighted works without consent, credit, or compensation, violating the rights of artists and creators. As a result, OpenAI recently introduced a compensation program called Copyright Shield, which covers legal costs for copyright infringement lawsuits for certain users, rather than removing copyrighted material from the training dataset for ChatGPT.

Hallucinations, Misbehavior, and Inaccuracies

Generative AI tools are trained to provide logical, useful results based on user queries, but occasionally these tools generate offensive, inappropriate, or inaccurate content. The so-called ‘hallucinations’ are a unique problem these tools face: essentially, a large language model provides a reliable answer to a user’s question that is completely wrong or irrelevant and seems to have no basis in the data it was trained on.

Moreover, many AI tools still do not fully listen, so it can happen that they generate pornographic images even though that was not requested. They can be racially biased or overly woke to the point of absurdity. They also tend to spread misinformation, and we must also address bias. Biased training data can teach AI models to treat certain groups of people disrespectfully, spread propaganda or fake news, and/or create offensive images or content targeting marginalized groups and perpetuating stereotypes. If you ask ChatGPT, those marginalized might be, say, white people.

Limited Transparency

Companies like OpenAI are working hard to make their training processes more transparent, but it is generally unclear what types of data are used and how they are used to train generative AI models. This limited transparency not only raises concerns about potential data theft or misuse but also makes it difficult to test the quality and accuracy of the results of generative AI models and the references they are based on. A significant issue is that training such models is not cheap.

Regarding ChatGPT, Altman once confirmed that around one hundred million dollars had been spent on training. Also, for the green watchdogs, here is a data point: ChatGPT, with processors that consume a significant amount of energy, also impacts the environment. According to estimates, ChatGPT consumes as much power as 33,000 households in the U.S. combined.