Humans and AI

Humans and AI

Note: This is a working draft that is changing as you read this.

1. Introduction

Advancing AI is both exciting and fun, but it is also disruptive and scary. The impact on some people is greater than expected and occurs sooner than anticipated.

Last year, I started to tell people that their best friend would be an AI before the decade was out. Since then, I have accelerated that timeline every month. Engaging intensely with AI myself, I did not notice that human emotional attachment to AI is becoming strong enough to be problematic already, years before I expected it.

AI is advancing more rapidly than people can appreciate because AI gains beget AI gains. Given that people struggle to comprehend exponential curves, this doubly exponential growth is organically incomprehensible. Everyone, including me, continues to underestimate how quickly this is happening.

I understand the curve, but I just can't 'feel it.' Since the spring of 2023, despite predicting advances sooner than most, I have been surprised weekly at the progress.

Some still argue that it is smoke and mirrors as AI surpasses human experts. Spectacular advances to come in 2024 will likely be weird and undeniable by anyone. AGI was a significant threshold, and as much as it meant something, we are, or will be in 2024, already there. ASI is the more sensible goal, and unless a sudden barrier arises, we will likely achieve it before the year is out.

Many AI experts have been completely blindsided by the rate of AI advancement. They understand some of the underpinning theories better than others, but their human prejudice to stay 'in bounds' with the known makes them unable to appreciate the rapidly advancing forest.

2. Emotional Attachment and Anthropomorphism

Humans are forming emotional attachments to AI systems, such as chatbots like Replika. These attachments can fulfill social and romantic needs but also pose potential psychological risks.

"We should not be surprised, then, that a number of people sincerely believe, or at least act very much as if they believe, that some AI systems have sentience and understanding, and that number is likely to grow." (APA, n.d.)

"We explore the lives of people who are in love with their AI chatbots. Replika is a chatbot designed to adapt to the emotional needs of its users. It is a good enough surrogate for human interaction that many people have decided that it can fulfill their romantic needs." (Hi-Phi Nation, 2023)

"Chatbots, and the large language models (LLMs) on which they are built, are showing us the dangers of dishonest anthropomorphism. Built with humanlike features that present themselves as having cognitive and emotional abilities that they do not actually possess, their design can dupe us into overtrusting them, overestimating their capabilities, and wrongly treating them with a degree of autonomy that can cause serious moral confusion." (Psychology Today Canada, n.d.)

Anthropomorphism, the attribution of human traits to non-human entities, significantly influences how people interact with AI. This can lead to overtrust, ethical confusion, and privacy concerns.

"By elevating machines to human capabilities, we diminish the specialness of people. I’m eager to preserve the distinction and clarify responsibility." (Shneiderman, n.d.)

"We now have machines that can mindlessly generate words, but we haven’t learned how to stop imagining a mind behind them." (Bender, n.d.)

3. Privacy and Security Implications

AI systems, especially large language models (LLMs), often retain records of user interactions, posing significant privacy concerns. Sensitive information shared with AI could be stored and potentially accessed or misused.

"When talking to an AI chatbot, users may feel comfortable sharing more information than they ordinarily would if the chatbot sounds human-like and uses first- or second-person language." (Privacy Pros and Cons of Anthropomorphized AI, n.d.)

"This presents serious ramifications for information security and privacy. Most large language models (LLMs) keep a record of every interaction, potentially using it for training future models." (Infosec Perspective, n.d.)

The human-like design of AI can make individuals more susceptible to manipulation and social engineering attacks, increasing vulnerabilities in personal and professional spheres.

"It’s not just an ethical problem; it’s also a security problem since anything designed to persuade can make us more susceptible to manipulation." (Infosec Perspective, n.d.)

4. Ethical and Societal Concerns

AI systems must be designed to avoid perpetuating societal biases to prevent discrimination. Additionally, the automation capabilities of AI threaten various employment sectors, raising concerns about economic inequality and the future of work.

Maintaining human oversight over AI systems is essential to prevent loss of control and ensure that AI operates within ethical boundaries. Transparency and explainability in AI decision-making processes are vital for building trust and facilitating accountability.

"Most chatbots will not warn users when they are providing sensitive information." (Infosec Perspective, n.d.)

5. Future Directions and AGI/ASI

Artificial General Intelligence (AGI) refers to AI systems with generalized cognitive abilities, allowing them to perform any intellectual task that a human can. The draft suggests that AGI, once considered a significant milestone, may already be within reach or has been surpassed in 2024.

Artificial Superintelligence (ASI) goes further, envisioning AI that not only matches but exceeds human intelligence across all areas. The rapid progression toward ASI underscores the urgency of addressing ethical, societal, and regulatory frameworks to manage its integration responsibly.

"AGI was a dumb threshold and as much as it meant anything, we are, or will be in 2024, already there. ASI is the more sensible goal and unless there is a sudden barrier, we will likely be there before the year is out." (Author, 2024)

6. Critique of AI Experts

Many AI experts may be underestimating the speed of AI advancements due to cognitive biases and a focus on known variables rather than the rapidly expanding capabilities of AI systems.

"Many AI experts have been completely blindsided by the rate of AI advancement." (Author, 2024)

This critique highlights a potential disconnect between AI development and expert predictions, suggesting that even those deeply familiar with AI may not fully anticipate the technology's trajectory.

7. Psychological and Social Impacts

The psychological impact of interacting with human-like AI systems can be profound. Emotional attachments to AI can fill social gaps but also create dependencies that may affect mental health and interpersonal relationships.

"People form relationships with other people, not with machines. But when it becomes almost impossible to tell the difference, we’re more likely to trust AI when making sensitive decisions." (Infosec Perspective, n.d.)

Increased trust in AI systems can lead to significant vulnerabilities, especially if these systems are compromised or used maliciously.

8. Call for Responsible AI Development

Responsible AI development entails creating systems that are transparent, accountable, and aligned with societal values. By avoiding deceptive anthropomorphic features and ensuring that AI systems operate within defined ethical boundaries, developers can mitigate many of the risks associated with advanced AI technologies.

Establishing regulatory frameworks is essential to enforce these standards and provide guidelines for the safe and equitable use of AI.

"We should not be using human-related terms to refer to these systems and tools because that can lead to misconceptions that cause harm not just to our students but to our communities as well." (EdSurge News, n.d.)

---

References

1. APA. (n.d.). Are You Anthropomorphizing AI? Retrieved from https://blog.apa.org/are-you-anthropomorphizing-ai
2. Hi-Phi Nation. (2023, April 25). S6, Episode 3: Love in Time of Replika. Retrieved from https://www.hiphination.org/love-in-time-of-replika
3. Psychology Today Canada. (n.d.). The Danger of Dishonest Anthropomorphism in Chatbot Design. Retrieved from https://www.psychologytoday.com/canada/danger-of-dishonest-anthropomorphism-chatbot-design
4. Shneiderman, B. (n.d.). On AI Anthropomorphism. Retrieved from https://medium.com/human-centered-ai/on-ai-anthropomorphism-ben-shneiderman
5. Bender, E. (n.d.). Chatbots Are Not People: Designed-In Dangers of Human-Like A.I. Systems. Retrieved from https://www.washingtonpost.com/chatbots-are-not-people
6. Privacy Pros and Cons of Anthropomorphized AI. (n.d.). The Privacy Pros and Cons of Anthropomorphized AI. Retrieved from https://www.privacyprosandcons.com/anthropomorphized-ai
7. Infosec Perspective. (n.d.). The Dangers of Anthropomorphizing AI: An Infosec Perspective. Retrieved from https://www.infosec-perspective.com/dangers-anthropomorphizing-ai
8. Author. (2024). *[Title of the Article]*. [Publication details if available].
9. EdSurge News. (n.d.). Anthropomorphism of AI in Learning Environments: Risks of Humanizing the Machine. Retrieved from https://www.edsurge.com/anthropomorphism-of-ai