Episode #6: Generative AI & Human Abstract Thinking
ChatGPT is in its infancy, having been released only six and a half months ago, isn’t it? Google Bard is even younger, May 10 was its beta’s official debut day, the day when I tested it with the Mental Flossing inquiry. With Bing Chat in the middle, they all are still under development. And there is a long journey ahead of them. Lots of hype right now, and lots of work ahead…
On June 8th, MIT reported exciting news “Our self-trained, 350-million-parameter entailment models, without human-generated labels, outperform supervised language models with 137 to 175 billion parameters.” The result of this research has significant implications for the LLM industry which has been actively looking for solutions to make scalable smaller LLMs that are much more accurate and energy efficient.
On the other hand, the debate about the possible AI-induced human extinction is about to become a cliche now. A healthy level of paranoia could save ourselves and our companies as we mentioned in Episode #3. However, being fearful could be counterproductive at many different levels. In case you want to get a grounded perspective on this subject, this article could be of some help.
Regardless, I have been curious about how much Abstract Thinking AI has been developing, which has a far more direct impact on our fate as humans, at least in the present moment.
From February to March, I spent quite a substantial amount of time crafting Episodes #2 and #3 from scratch and only used ChatGPT for a small amount of reference checking: 1) I didn’t think that GenAI was good enough yet to do human-level abstract thinking; 2) GenAI extracts human thoughts from the past data pools but can’t relate to the current affairs yet; 3) GenAI can’t do cross reference and gather analogies from a broad range of history, disciplines, literature, and other fields, just yet.
Although, in May, when Google Bard was released officially, I decided to experiment with GenAI for my first draft — to test out how far can it go in terms of Abstract Thinking. It would be interesting to compare these three: GPT-4, Bing Chat(creative), and Bard. This comparison could help us better understand the capabilities of these language models and identify areas where they can be improved.
If GenAI can do all the heavy lifting, let it be… It’s time to embrace this new paradigm wholeheartedly with a proactive open mind, as long as we are using it ethically and responsibly. The bottom line is about building trust between GenAI and humans, which wouldn’t happen overnight. It’s going to take numerous trials and errors, and lots of technical improvements along the way. Very similar to building trust between humans — trust wouldn’t come by itself, and solid trust comes from learning and accepting each other’s differences, out of profound respect and gratitude. Oftentimes, it comes with lots of pain and sacrifices.
Before we dive deeper, I asked GenAI to share some highlights about its own history and its relative relationship with traditional AI. Most fear comes from a lack of understanding of the truth. The best way to get around is to educate ourselves as much as we can.
Because Bing Chat(creative) is the equivalent of GPT-4, we are going to pick just one of these two. From the results shown below, one can see that Google Bard understands diagrams better than Bing Chat. The diagram link provided by Big Chat didn’t work.
The prompt: please make a diagram comparing the history of traditional AI and generative AI. thank you.
Google Bard:
Sure, here is a diagram comparing the history of traditional AI and generative AI:
Traditional AI is a field of computer science that seeks to create intelligent agents, which are systems that can reason, learn, and act autonomously. Traditional AI systems are typically rule-based, meaning that they are programmed with a set of rules that they use to make decisions. Traditional AI has had some successes, such as in the development of chess-playing programs and medical diagnosis systems. However, traditional AI has also faced challenges, such as the difficulty of programming systems to handle complex real-world situations.
Generative AI is a newer field of AI that seeks to create systems that can generate new ideas and concepts. Generative AI systems typically use statistical models to learn from data and then use that knowledge to generate new outputs. Generative AI has the potential to revolutionize many industries, such as the media, entertainment, and healthcare industries.
The following table provides a more detailed comparison of traditional AI and generative AI:
Overall, traditional AI and generative AI are two different approaches to artificial intelligence. Traditional AI is focused on rule-based systems that can solve well-defined problems, while generative AI is focused on statistical models that can generate new ideas and concepts. Both approaches have their own strengths and weaknesses, and they are likely to continue to be developed and used in the future.
Bing Chat (creative):
Sure, I can make a diagram comparing the history of traditional AI and generative AI. Here is the diagram:
The diagram shows some of the major milestones and achievements in the history of traditional AI and generative AI. Traditional AI refers to the branch of AI that focuses on creating systems that can perform specific tasks or solve specific problems, such as playing chess, recognizing faces, or translating languages. Generative AI refers to the branch of AI that focuses on creating systems that can generate new content or data, such as text, images, or music.
The diagram shows that traditional AI has a longer history than generative AI, dating back to the 1940s when the first electronic computers were invented. Some of the early examples of traditional AI include the Logic Theorist (1956), which could prove mathematical theorems; ELIZA (1966), which could simulate a psychotherapist; and SHRDLU (1970), which could manipulate blocks in a virtual world.
Generative AI emerged later, in the 1980s and 1990s, with the development of neural networks and genetic algorithms. Some of the early examples of generative AI include ALICE (1983), which could generate music based on rules; AARON (1987), which could paint original artworks; and Tierra (1991), which could simulate artificial life forms.
The diagram also shows that both branches of AI have experienced rapid progress and innovation in the 21st century, especially with the advent of deep learning and big data. Some of the recent examples of traditional AI include AlphaGo (2016), which could beat human champions at Go; FaceNet (2015), which could achieve near-perfect face recognition; and GPT-2 (2019), which could generate coherent text based on prompts.
Some of the recent examples of generative AI include DeepDream (2015), which could create psychedelic images based on neural network activations; WaveNet (2016), which could generate realistic speech and music; and DALL-E (2021), which could create images from natural language descriptions.
Source: Conversation with Bing, 6/13/2023
(1) The brief history of artificial intelligence: The world has changed …. https://ourworldindata.org/brief-history-of-ai.
(2) The History of Artificial Intelligence — Science in the News. https://sitn.hms.harvard.edu/flash/2017/history-artificial-intelligence/.
(3) A Complete History of Artificial Intelligence — G2. https://www.g2.com/articles/history-of-artificial-intelligence.
(4) Traditional AI vs. Modern AI.. The evolution of Artificial… | by Awais …. https://towardsdatascience.com/traditional-ai-vs-modern-ai-5117b469a0c9.
(5) History of artificial intelligence | Codebots. https://codebots.com/artificial-intelligence/history-of-artificial-intelligence.
The second prompt:
Please write an essay comparing Generative AI and Human Abstract Thinking, in the style of a Time magazine journalist.
The answers from Google Bard and Bing Chat(creative) are quite different, as expected. Although they seem to complement each other. Don’t know if they can spy on each other and use my chat history simultaneously.
What do you think? Are we ready to let GenAI take over the writing business? Will all magazines disappear in the near future? Or will this challenge inspire us to raise our standards? To compete with GenAI, we will have to outperform it not just by a margin, but by a factor of 5, 10, or higher… Would that be a good challenge for us to find ways to boost our human Abstract Thinking?
Google Bard:
Sure, here is an essay comparing Generative AI and Human Abstract Thinking, in the style of a Time magazine journalist:
Generative AI vs. Human Abstract Thinking
The development of generative AI has led to a renewed debate about the nature of human intelligence. Some experts believe that generative AI will eventually surpass human abstract thinking, while others believe that humans will always retain a unique advantage.
Generative AI is a type of artificial intelligence that can create new content, such as text, images, and music. It does this by learning from a large dataset of existing content. For example, a generative AI system could be trained on a dataset of books to learn how to write new stories.
Human abstract thinking is the ability to think about things that are not immediately present or tangible. It is what allows us to solve problems, create new ideas, and understand the world around us.
There are some similarities between generative AI and human abstract thinking. Both can generate new ideas and concepts. However, there are also some important differences. Generative AI is limited by the data that it is trained on. It can only generate new content that is similar to the content that it has seen before. Human abstract thinking, on the other hand, is not limited by data. We can come up with new ideas that are completely original.
Another important difference is that generative AI is not creative. It can only generate content that is already possible. Human abstract thinking, on the other hand, is creative. We can come up with new ideas that are not possible in the real world.
Overall, generative AI and human abstract thinking are two different things. Generative AI can be a powerful tool for generating new content, but it is not a replacement for human abstract thinking. Human abstract thinking is what makes us human, and it is what allows us to create a better world.
The Future of Generative AI and Human Abstract Thinking
The development of generative AI is still in its early stages, but it has the potential to revolutionize the way we live and work. Generative AI could be used to create new forms of art, music, and literature. It could also be used to develop new educational tools and new ways to solve problems.
However, there are also some potential risks associated with the development of generative AI. For example, generative AI could be used to create harmful or misleading content. Additionally, generative AI could be used to create autonomous systems that could pose a threat to human safety.
It is important to remember that generative AI is a tool. It can be used for good or for evil. It is up to us to decide how we will use this technology.
If we use generative AI wisely, it has the potential to make the world a better place. It can help us to solve problems, to create new ideas, and to understand the world around us. However, if we use generative AI unwisely, it could pose a serious threat to humanity.
The future of generative AI and human abstract thinking is uncertain. However, it is clear that these two technologies are going to play a major role in the future of our world. It is up to us to decide how we will use these technologies to shape our future.
Bing Chat(creative):
Here is a possible essay:
Generative AI and Human Abstract Thinking: A Comparison
Generative AI is a type of artificial intelligence (AI) system that can create new content, such as text, images, music, or code, based on prompts or data. Human abstract thinking is the ability to understand and reason about concepts that are not directly observable, such as freedom, love, or wisdom. Both generative AI and human abstract thinking involve going beyond the concrete and the literal, and finding connections and patterns that are not obvious. However, there are also significant differences between them in terms of how they work, what they can do, and what challenges they face.
How They Work
Generative AI systems work by learning from large amounts of data, such as text corpora, images, or audio files, using machine learning algorithms. These algorithms can be unsupervised, meaning they do not require labels or feedback, or self-supervised, meaning they use part of the data as labels or feedback. For example, a generative AI system that creates text can use the previous words or sentences as labels for predicting the next word or sentence. Generative AI systems can use different architectures or models to generate content, such as variational autoencoders (VAEs), generative adversarial networks (GANs), or transformers. These models can capture the statistical patterns and structure of the data, and then produce new data that has similar characteristics.
Human abstract thinking works by using mental representations, such as symbols, images, words, or schemas, to capture the meaning and essence of concepts that are not directly observable. Human abstract thinking can use different cognitive processes to manipulate these representations, such as analogy, metaphor, inference, deduction, induction, or synthesis. These processes can help humans to discover new relationships and insights among concepts, and to apply them to different domains and contexts.
What They Can Do
Generative AI systems can do a variety of tasks that require creativity and innovation, such as:
- Creating art, music, poetry, or stories
- Generating captions, summaries, headlines, or slogans
- Writing essays, articles, reports, or reviews
- Designing logos, icons, fonts, or layouts
- Developing software code, algorithms, or applications
- Optimizing business processes, workflows, or strategies
- Simulating scenarios, environments, or outcomes
Human abstract thinking can do a variety of tasks that require understanding and reasoning about complex concepts and situations. Some examples are:
- Solving problems that have multiple solutions or no clear solution
- Making decisions that involve trade-offs or uncertainties
- Evaluating arguments that have different perspectives or assumptions
- Forming theories that explain phenomena or events
- Developing hypotheses that test predictions or implications
- Creating models that simplify or generalize reality
- Learning new concepts that extend existing knowledge
What Challenges They Face
Generative AI systems face several challenges that limit their performance and applicability. Some of these are:
- Data quality and quantity: Generative AI systems depend on the availability and reliability of the data they learn from. If the data is noisy, Incomplete, biased, or outdated, the systems may produce inaccurate, inconsistent, unfair, or irrelevant content.
- Evaluation and validation: Generative AI systems are difficult to evaluate and validate because there is no objective or universal criterion for measuring their quality or usefulness. Different users may have different preferences, expectations, or goals for the content they want to generate or consume.
- Ethics and responsibility: Generative AI systems raise ethical and social issues because they can affect human values, beliefs, and behaviors. For example, they can create fake news or deepfakes that can deceive or manipulate people; they can infringe on intellectual property rights or privacy rights of the original creators or owners of the data; they can replace human workers or artists in some domains or tasks.
Human abstract thinking also faces several challenges that hinder its development and application. Some of these are:
- Cognitive biases and heuristics: Human abstract thinking is prone to errors and fallacies because humans tend to rely on shortcuts or rules of thumb that simplify complex situations but may lead to inaccurate or irrational conclusions. For example, humans may confirm their existing beliefs rather than seek disconfirming evidence; they may overestimate their own abilities or knowledge rather than admit their limitations; they may follow the majority opinion rather than form their own judgment.
- Emotions and motivations: Human abstract thinking is influenced by emotions and motivations that may affect its objectivity and validity. For example, humans may feel anxious or frustrated when faced with difficult problems; they may feel happy or proud when they find a solution or create something new; they may feel curious or bored when they encounter new concepts or situations.
- Education and culture: Human abstract thinking is shaped by education and culture that may affect its diversity and richness. For example, humans may learn different ways of thinking and expressing abstract concepts depending on their language, literacy, or discipline; they may encounter different sources and examples of abstract thinking depending on their exposure, access, or interest; they may adopt different values and norms of abstract thinking depending on their society, community, or group.
Conclusion
Generative AI and human abstract thinking are both important and fascinating aspects of intelligence that enable us to create and understand new and complex content. They have similarities and differences in how they work, what they can do, and what challenges they face. By comparing and contrasting them, we can gain a better appreciation of their strengths and weaknesses, and their potential and limitations. We can also explore how they can complement and enhance each other, and how they can collaborate and coexist in harmony.
Source: Conversation with Bing, 6/14/2023
(1) Generative artificial intelligence — Wikipedia. https://en.wikipedia.org/wiki/Generative_artificial_intelligence.
(2) What is ChatGPT, DALL-E, and generative AI? | McKinsey. https://www.mckinsey.com/featured-insights/mckinsey-explainers/what-is-generative-ai.
(3) Generative AI defined: How it works, benefits and dangers — TechRepublic. https://www.techrepublic.com/article/what-is-generative-ai/.
(4) Understanding Abstract Thinking: Development, Benefits & More — Healthline. https://www.healthline.com/health/abstract-thinking.
(5) https://bing.com/search?q=Human+Abstracting+Thinking.
(6) Abstract Thinking: Definition, Examples, Uses, and Tips — Verywell Mind. https://www.verywellmind.com/what-is-abstract-reasoning-5181522.
(7) Human need to cultivate abstract thinking — ResearchGate. https://www.researchgate.net/profile/Mahmood-Bahrani/publication/350630960_Human_need_to_cultivate_abstract_thinking/links/606ac46da6fdccad3f71a379/Human-need-to-cultivate-abstract-thinking.pdf.
NOTE: Thanks to Google Bard for some great editing suggestions on human Karolyn’s native writing.
