AI - The Opportunities and Risks

Despite the promise of AI, there are serious challenges to overcome. Data protection, ethical concerns and the potential for abuse are just some of them. It is essential that we take a serious look at not only the opportunities, but also the risks of AI, and take appropriate measures to address them. But first, let’s look at the potential of artificial intelligence from a mathematical perspective. Dr. Julian Feinauer, founder of Pragmatic Industries and Pragmatic Minds, explained very clearly at the event “Artificial Intelligence – Opportunity or Challenge?” initiated by Dr. Natalie Pfau-Weller in Kirchheim unter Teck in April 2024, what artificial intelligence actually is from a mathematical perspective, what its potential is and what artificial intelligence has to do with energy.

Linear Regression and Artificial Intelligence

Most people are familiar with the concept of linear regression from their 8th grade curriculum, which involves the simple formula Ax+B. However, in the world of artificial intelligence (AI), this concept expands exponentially. Instead of just two axes of coordinates, as in linear regression, we have countless axes. The result is no longer a simple line, but a dynamic, moving surface. With models that have over 400 billion or even 700 billion parameters, it becomes clear just how complex and powerful AI can be. From a mathematical point of view, this is the essence of AI.

AI application examples

Artificial intelligence offers endless opportunities for progress and innovation. From automating work processes to developing personalized solutions, AI has the potential to transform almost every aspect of life. With the ability to recognize patterns, make decisions and solve complex problems, AI is paving the way for all kinds of new possibilities. Here are some of the examples Dr. Feinauer used in his presentation to illustrate how AI is already being used today.

Image generation with artificial intelligence

Generated images have attracted a lot of attention recently, especially in relation to the question of which of these images are real and which are not. It is fascinating to see how AI models such as VASA A1 are able to generate complex visual and auditory content from a single photo and a soundtrack. But how exactly does this work?

If a very noisy image is presented to an AI for denoising and the AI is repeatedly instructed to enhance the image, it will most likely recognise and filter out certain patterns. For example, a cat in a very noisy image can be uncovered by repeatedly instructing the AI to denoise the image. This demonstrates the AI’s amazing ability to recognise and reconstruct complex patterns.

The technology behind AI-based image generation is developing rapidly. The next step in image generation is AI video generation. Just recently, Microsoft introduced VASA A1, one such tool.

AI video generation from photos and audio: impressive progress

AI conquers video editing

With tools such as VASA A1, it is now possible to create a realistic animated video from a single photo and a soundtrack. The results are amazing. The AI can animate faces realistically and even synchronise lip movements with the audio track. The result is a smooth and natural video that is almost indistinguishable from real film footage at first glance. Despite this progress, there is still room for improvement. In some cases, small errors can occur, such as slightly slipped teeth when speaking. The AI-generated environment also still looks a little artificial at times.

Basically, AI video generation follows the same principle as image generation and can be seen as a kind of de-noising. Based on the photo and the audio track, the AI “removes” blurriness and inaccuracies and fills the gaps with realistic details.

Development is progressing rapidly. It can be assumed that AI video generators will deliver even more precise and realistic results in the near future. This could fundamentally change the way we create and consume videos.

Potential fields of application

There are many potential applications for AI-generated video. For example, they could be used in film production, advertising, education or even social media.

Ethical concerns

However, the new technology also raises ethical concerns. For example, AI video generation could be misused to create deepfakes, i.e. videos in which people are portrayed in a manipulated manner. It is therefore important that this technology is used responsibly.

AI video generation is still at an early stage, but offers enormous potential. As the technology develops, we can expect even more amazing and realistic videos in the future. However, it is important to be aware of the ethical concerns and use the technology responsibly.

Language models with artificial intelligence

Another area that is developing rapidly is large language models such as ChatGPT. These models are based on the principle of probability calculation, where each word is assigned a specific number and the probability of the next word is calculated based on the previous context. By training with huge data sets, the AI can learn to generate realistic-looking texts and even respond to questions. However, Dr. Feinauer shows in his presentation that the answers are not always accurate using an application example with ChatGPT. He asked ChatGPT for Dr Natalie Pfau-Weller, CDU member of parliament for the Kirchheim unter Teck constituency, and received a humorous answer.

In the world of artificial intelligence, Dr. Pfau-Weller is a hit star. She has even released resounding hits and delights her fans on tour.

This apparent reality reveals a crucial weakness of language models. The use of tools such as ChatGPT or Gemini can undoubtedly make life easier by automating complex tasks and saving time. However, it is essential to always use common sense to check whether the results generated correspond to the actual truth. Despite these weaknesses, language models can be useful tools if they are used responsibly. However, it is important to be aware of their limitations and to recognise their weaknesses.

The weaknesses of language models

• Prejudice and discrimination: Texts can be discriminatory or offensive.
• Lack of factual accuracy: Texts can be grammatically correct but not correspond to reality.
• Manipulability: Texts may contain fake news or propaganda.
• Lack of common sense: Texts may be correct but not make sense.
• Lack of creativity: Texts often resemble training data and are not really creative.

Additional limitations:

• Size and computing power: High demand on computing power for training and use.
• Data availability: Large amounts of data required for training in new languages or domains.
• Cost: Training and utilisation can be expensive.
• The energy demand resulting from the high computing power requirements of AI applications.

Where do we actually get the energy for AI applications?

A single ChatGPT query consumes around 3000 watt-hours of energy, the equivalent of 21 cups of coffee or driving around 1.4 kilometres in an electric car. It is interesting to note that the energy consumption of a single ChatGPT query is equivalent to the energy consumption of 1000 Google searches. Assuming that each of ChatGPT’s 100 million active users makes only one query per month, this would correspond to a total energy consumption of 2.1 billion cups of coffee.

According to a study by Joule magazine, the energy consumption of AI-related computing could exceed 134 terawatt-hours by 2027. 134 TWh. To put this in perspective, all German households together consume just over 100 TWh per year.

When we talk about training AI models, the energy requirements of these models often exceed the energy consumption of many households. Everyone should be aware of this when using AI models. It is more important than ever that we find sustainable ways to meet this immense energy demand, which will only increase in the coming years. AI is here to stay, and models will continue to be developed and used. That takes energy.

Ethical issues and social impact

The introduction of AI also raises important ethical questions, particularly in relation to its impact on our society. Issues of justice, equality and transparency need to be carefully considered to ensure that AI is used for the benefit of all and not to the detriment of some. When we consider the likely response to the question of how authoritarian states might use these developments, we conclude that it is of the highest priority to monitor ethical responsibility in the use of AI models.

Regulation for AI applications – EU AI ACT

With the growing influence and proliferation of artificial intelligence (AI) in various areas of the economy and daily life, the need for appropriate regulation to make the use of this technology safer and more ethically responsible is also increasing. In this context, the European Union (EU) has recently proposed the AI Act, a landmark legislation that aims to establish clear rules for the development, use and monitoring of AI applications. The AI Act is currently at an advanced stage of the legislative process. Extensive discussions, consultations and assessments have already been conducted to scrutinize the various aspects of the Act and ensure that it meets the needs and requirements of both industry and society. Although the exact timetable for the adoption of the AI Act has not yet been finalized, there is every indication that the EU Member States and the relevant institutions are determined to move the process forward and improve the regulation of AI applications in the foreseeable future.

A summary of the main contents of the EU AI Act

1. Risk-based classification: AI systems are categorised based on their risk potential, with strict requirements for high-risk systems.
2. Data protection and human rights: The AI Act protects privacy and personal rights by preventing invasive or discriminatory use of AI technologies.
3. Transparency requirements: Clear transparency requirements apply to certain AI systems, such as chatbots, to educate users about AI interactions.
4. Technical requirements: The act requires the development of AI systems that respect human autonomy and avoid harm.
5. Prohibited practices: Certain applications that could undermine the free will of users are explicitly prohibited

Opportunities

1. Process optimization: AI enables a significant increase in process efficiency through automation and more precise control of business processes.
2. Data-based decision-making: AI helps companies gain valuable insights from large amounts of data, enabling them to make informed decisions and increase their competitiveness.
3. Personalized customer approach: AI enables companies to develop customized marketing strategies and products that are better tailored to customers’ needs and preferences.
4. More efficient supply chain management: AI helps optimize supply chains by improving forecasting and tracking of deliveries, resulting in lower costs and faster delivery times.
5. Increasing operational safety: AI can help to increase safety in the workplace by monitoring and predicting potential hazards, especially in the manufacturing industry.

Risks

1. Data protection concerns: AI systems that process large amounts of company and customer data harbor the risk of data breaches that can jeopardize customer trust and compliance with legal regulations.
2. Decision bias: If AI systems are trained on the basis of biased or incomplete data, they can make erroneous or discriminatory decisions that pose legal and reputational risks.
3. Security vulnerabilities: AI-based systems can be the target of cyber attacks that jeopardize not only IT security, but also the physical security of automated or autonomous systems.
4. Overdependance on technology: The frequent use of AI systems can lead to strong dependency and the associated loss of critical thinking and decision-making skills, which is particularly risky in crisis situations.
5. Regulatory compliance: The rapid development of AI technology can lead to uncertainties in complying with existing and future regulatory requirements, which poses legal challenges for companies.