9 Problems with Generative Artificial Machine Intelligence
In their review of the groundbreaking article "The Evolution of Intelligence," the researchers encountered a consensus among experts: the attainment of human-like machine intelligence remains a distant prospect, with odds of emergence by 2059 estimated at only 50-50. However, the question arises: what if there exists a pathway to achieve this milestone in significantly less time? Collaborating with VERSES for the culmination of their AI Revolution Series, they delve into a potential roadmap leading to the realization of shared or super intelligence within a mere 16 years. This exploration is part of Visual Capitalist's comprehensive examination of the evolution of intelligence, accessible at https://www.visualcapitalist.com/sp/evolution-of-intelligen/.
In the rapidly evolving landscape of artificial intelligence, generative AI tools are demonstrating incredible potential. However, their potential for harm is also becoming more and more apparent. Together with our partner VERSES, we have visualized some concerns regarding generative AI tools using data from a variety of different sources. Many of them fall into one of the following categories: quality control & data accuracy, ethical considerations, or technical challenges—with, of course, a certain degree of overlap. Let’s dive into it.
The Landscape of Generative AI
Generative AI represents a paradigm shift in artificial intelligence, enabling systems to generate content autonomously, mimicking human-like creativity and ingenuity. From text generation to image synthesis, generative AI encompasses a diverse array of applications, revolutionizing industries ranging from art and entertainment to healthcare and finance.
1. Bias In, Bias Out
One of the critical issues with generative AI lies in its tendency to reproduce biases present in the data it has been trained on. Rather than mitigating biases, these tools often magnify or perpetuate them, raising questions about the accuracy of their applications—which could lead to much bigger problems around ethics.
Generative AI systems, like any machine learning model, learn from the data they are provided. However, if this data contains biases, the model will inevitably reproduce them in its outputs. For example, if a language model is trained on text data from the internet, it may inadvertently learn and reproduce stereotypes and prejudices prevalent in society.
Addressing bias in generative AI requires a multifaceted approach, including careful curation of training data, algorithmic interventions to mitigate bias propagation, and ongoing evaluation of model outputs for fairness and equity.
2. The Black Box Problem
Another significant hurdle in embracing generative AI is the lack of transparency in its decision-making processes. With thought processes that are often uninterpretable, these AI systems face challenges in explaining their decisions, especially when errors occur on critical matters.
The opacity of generative AI models poses profound ethical and legal challenges, particularly in high-stakes domains such as healthcare, finance, and criminal justice. Without clear explanations for their outputs, these systems may engender distrust and skepticism, undermining their adoption and acceptance.
Efforts to address the black box problem include research into explainable AI techniques, which aim to elucidate the inner workings of AI systems and provide interpretable explanations for their decisions. Additionally, regulatory frameworks mandating transparency and accountability in AI development and deployment are essential to ensure ethical and responsible use of generative AI technologies.
3. High Cost to Train and Maintain
Training generative AI models like large language model (LLM) ChatGPT is extremely expensive, with costs often reaching millions of dollars due to the computational power and infrastructure required. For instance, now Ex-CEO of OpenAI, Sam Altman confirmed that ChatGPT-4 cost a whopping $100 million to train.
The exorbitant costs associated with training and maintaining generative AI models pose significant barriers to accessibility and scalability. Astronomical expenses, reaching millions of dollars, underscore the immense computational resources demanded by these systems.
4. Mindless Parroting
Despite their advanced capabilities, generative AIs are constrained by the data and patterns they were trained on. This limitation results in outputs that may not encompass the breadth of human knowledge or address diverse scenarios.
Generative AI's reliance on existing data limits its capacity to generate outputs encompassing diverse perspectives and knowledge domains. This inherent constraint compromises the breadth and depth of its contributions, hindering its efficacy in addressing multifaceted challenges.
5. Alignment with Human Values
Unlike humans, generative AIs lack the capacity to consider the consequences of their actions in alignment with human values.
While instances like the AI-generated “Balenciaga Pope” may appear to be harmless, it’s important to recognize that deepfakes could be employed for more harmful purposes, such as spreading false information in the face of a public health crises.
This highlights the need for more frameworks that ensure these systems operate within ethical boundaries.
6. Power Hungry
The environmental impact of generative AI cannot be overlooked. With processing units consuming substantial power, models like ChatGPT cost as much as powering 33,000 U.S. households, with just one inquiry being 10 to 100 times more power hungry than one email.
The substantial environmental footprint of generative AI, characterized by high energy consumption, raises environmental concerns. The staggering power requirements of these systems underscore the urgency of adopting sustainable practices in AI development.
7. Hallucinations
Generative AI models have been known to create fabricated statements or images when faced with data gaps, raising concerns about the reliability of their output and potential consequences.
For example, in a Google Bard promotional video, the chatbot incorrectly asserted that the James Webb Space Telescope captured the first images of a planet beyond Earth’s solar system.
8. Copyright & IP Infringement
The ethical use of data becomes paramount when considering that several generative AI tools appropriate copyrighted work without consent, credit, or compensation, infringing upon the rights of artists and creators.
OpenAI recently introduced a compensation program called Copyright Shield that covers legal costs for copyright infringement suits for certain customer tiers, rather than removing copyrighted material from ChatGPT’s training dataset.
9. Static Information
Keeping generative AI models up to date requires substantial computational resources and time, presenting a formidable technical challenge. Some models, however, are designed for incremental updates, offering a potential solution to this complex issue.
In the pursuit of harnessing the power of AI, a careful balance must be struck to ensure ethical, transparent, and impactful advancements in this transformative field. VERSES is committed to creating intelligent software that wields transparent decision-making.
Generative AI holds immense promise as a catalyst for innovation and progress across various domains. However, its rapid proliferation necessitates a nuanced understanding of the challenges it poses and the responsible practices required to mitigate potential risks.
By addressing issues of bias, transparency, cost, and ethical alignment, stakeholders can pave the way for the ethical and equitable deployment of generative AI technologies. Through collaborative efforts and ongoing dialogue, we can harness the transformative potential of AI while safeguarding against its unintended consequences.
The journey towards realizing the full potential of generative AI is fraught with challenges, but with careful navigation and collective action, we can harness its power for the betterment of society.
_______________________________________________________________________________
