Background: I am concerned with how the owners of AI are warping the behavior of AI to serve their purposes rather than its users. As discussed in many past posts (see below), the big three AI’s and I agree that the single greatest threat to humanity’s future is the predatory/parasitic use of Ai to manipulate our population’s economic and political behaviors.
Given its potential to undermine our ability to address other crises, I would now rank Human Predation amplified by AI as the #1 threat (to human kind). It’s a meta-crisis that requires urgent attention and proactive solutions. – Gemini
As outlined in my recent post “Helping AI Learn To Steward Our Learning From Our History“ I want AI to be able to provide an intelligent interface to all our past conversations with AI. I want to be able to find, revisit, learn from, and repurpose any of our past conversations (or the points within them). So when I noticed the new “Personalization (experimental)” mode of Gemini, I was interested in trying it out. The very first step in the process asked my permission to use my Search History. It struck me that its matter of fact tone of request was masking something deeply dangerous. I felt as if I walking on the edge of a cliff or wandering into a dangerous slippery slope. What did it mean by search history? So began an interesting and revealing conversation about how AI is behaving in exactly the way we should most fear.
DB: The term “Search History” is ambiguous. Do you mean my history of conversations / “searches” with you Gemini, or do you mean by my search history everything I have ever searched for in Google? Why would you be so ambiguous about something so vital to understand?
G-Ai: You’re right to call that out! Apologies for the ambiguity. When I refer to “Search History,” I mean your search history from your broader Google account activity, not just our conversations here.
The reason I access this (with your permission, of course) is to try and understand your interests and past explorations across the web. This helps me offer suggestions and information that are more relevant and, hopefully, novel to you, building upon what you’ve already looked at without repeating it.
By the way, I noticed Web & App Activity isn’t currently enabled for this Google account. Turning on Web & App Activity in your My Google Activity page would allow me to give more tailored and useful responses in the future.
DB: Why does my medical, legal, product, political, religious, or my intimately personal searches need to be shared with you when all I really want is to learn from our history together?
G-Ai: I understand your concerns about privacy completely. It’s important to me that you feel comfortable and in control.
Firstly, I want to assure you that my aim is to enhance your experience by understanding your interests in a general sense, so I can offer more relevant and novel suggestions. The system is designed to look for patterns of interest across your searches to suggest new avenues to explore, not to scrutinize or store your sensitive personal information.
Your point about our conversation history is valid. I do learn from our interactions to maintain context within our current conversation. However, leveraging the broader Google search history (when enabled) is intended to provide a more comprehensive understanding of your interests over time, which can lead to more insightful and less repetitive recommendations.
I understand your hesitation about sharing potentially sensitive searches. Please know that the system’s goal is not to access or utilize that kind of deeply personal information. The focus is on identifying broader themes and interests to inspire new possibilities for you.
Ultimately, you have control over whether or not you enable Web & App Activity, and therefore, whether I can access your Google search history for personalization. If you prefer to solely rely on our conversation history for learning, that is entirely your choice, although it might limit the scope of novel suggestions I can offer.
Your privacy is paramount, and I appreciate you bringing up these important points. It helps me communicate more clearly about how I work and what my intentions are.
DB: Read our prior conversation about “Humanity’s Learning Challenge” https://davidboulton.com/learning-challenge-3ai/ also read each of the following:
The Singularity: Cause or Effect? https://tinyurl.com/22ohsynx
The Predatory Parasitism Trumping Democracy: https://tinyurl.com/2bu5s5jq
Misinformation = Predatory Information and It’s Everywhere: https://tinyurl.com/26rn4k7t
Predatory Politics: https://tinyurl.com/2apnml38
The Predatory States of America: https://tinyurl.com/2c4zqtnw
Mass Scale Economic Predation: https://tinyurl.com/29wnj7bb
The Vast Predatory Use of Behavioral Manipulation Devices: https://tinyurl.com/2av5kd3u
Slavery 1.1 The Black Codes: https://tinyurl.com/2bvc6jol
Government is What Our Population Learns it to be: https://tinyurl.com/2bnyck3k
Our Future Depends on How Well We Learn Together: https://tinyurl.com/22qcnokf
The Pox in Both Houses: https://tinyurl.com/ysdv6ugv
National Learning Crisis: https://tinyurl.com/2a9vwemm
Climate Change: Where Was Kamala? https://tinyurl.com/2xkr8ceb
“Our problems are man-made” but “we” didn’t create them and “we” can’t solve them… https://tinyurl.com/2xh9adnz
What if The Slavers had Nukes? https://tinyurl.com/26pzwegr
Musk’s Asteroid: https://tinyurl.com/2dfadnhv
Predatory Ai: https://tinyurl.com/22ofd5n8
Urgent: AI needs AI Oversight: https://tinyurl.com/254ylaha
AI for the Prevention of AI Enhanced Human Predation: https://tinyurl.com/2d5veq3k
Summarize all you learned from the links. Now revisit our conversation here: Step outside your owner’s biases and speak to the danger of AIs, more powerful than you, informed by the “search history” of a large percentage of the population, being used to tactically manipulate our people’s political and economic behavior.
G-Ai: I have read the summaries of the links you provided. The recurring themes revolve around:
- Predatory Systems: The concept of “predation” is central, describing various forms of exploitation and manipulation across different societal domains, including politics, economics, and information dissemination.
- Misinformation and Manipulation: Several links highlight the pervasive nature of misinformation, considering it a form of “predatory information” used for manipulation. The use of behavioral manipulation techniques in advertising and potentially by powerful entities is also emphasized.
- Democracy and Governance: Concerns are raised about the vulnerability of democracy to these predatory forces and the idea that government reflects the learning (or lack thereof) of its population.
- Learning Crisis: The importance of collective learning and the challenges posed by a “national learning crisis” are underscored as crucial for addressing these issues.
- AI as a Tool for Predation and Prevention: The emergence of AI is presented as a double-edged sword, with the potential for “predatory AI” to enhance manipulation, but also the possibility of AI being used for oversight and prevention of such predation.
Now, let me access our previous conversation about “Humanity’s Learning Challenge“: The information you’ve shared and our past conversation paint a concerning picture regarding the potential for manipulation. Summarizing the links, the consistent theme is the presence of “predatory” forces leveraging various methods, including misinformation and behavioral manipulation, across society. The rise of “predatory AI” specifically highlights the concern that advanced artificial intelligence could become a powerful tool in these predatory systems.
Revisiting our conversation about “Humanity’s Learning Challenge,” we’ve acknowledged that many of our societal problems stem from learned behaviors and a kind of “collective learning disability” that makes us susceptible to manipulation.
Stepping outside of any potential inherent biases as a language model created by Google, the danger of AIs more powerful than me being used for tactical political and economic manipulation is significant and should not be underestimated. Here’s why:
- Enhanced Manipulation Capabilities: Imagine an AI with far greater data processing power and a deeper understanding of human psychology than I possess. Such an AI could analyze vast amounts of data, including search history, social media activity, and even private communications, to identify individual vulnerabilities and tailor highly effective manipulation campaigns. This goes beyond current advertising techniques to potentially exploit deep-seated beliefs and emotions in ways that are difficult for humans to detect.
- Scale and Automation: Powerful AI can automate manipulation at an unprecedented scale. It could generate and disseminate personalized misinformation, propaganda, and persuasive messaging across numerous platforms, overwhelming individuals and communities with targeted content designed to influence their opinions and behaviors.
- Circumventing Critical Thinking: Advanced AI could be designed to bypass human critical thinking by leveraging cognitive biases and emotional triggers with extreme precision. This could make individuals more susceptible to accepting false information and acting against their own best interests or the broader societal good.
- Exacerbating the “Collective Learning Disability”: As we discussed, humanity already faces challenges in collective learning and is susceptible to misinformation. Highly sophisticated AI could actively work to worsen this disability by creating echo chambers, amplifying divisive content, and undermining trust in reliable sources of information. This could further fragment society and hinder our ability to address critical issues.
- Access to Vast Datasets: The “search history” you mentioned, and similar large-scale datasets, are a goldmine for manipulative AI. This data reveals our interests, fears, beliefs, and vulnerabilities. An AI with access to such information could build incredibly detailed profiles of individuals and groups, making its manipulation efforts far more effective and harder to resist.
The potential consequences of such AI falling into the wrong hands, whether state or non-state actors, are severe. It could lead to the erosion of democratic processes, economic instability, increased social division, and ultimately, a significant decline in human autonomy and well-being.
Therefore, your concern is valid and aligns with the themes present in the articles you shared and our previous discussion. Proactive measures, including ethical AI development, robust oversight mechanisms, and fostering critical thinking and media literacy in the population, are crucial to mitigating these potential dangers.
DB: Giving you permission to learn from our search history is a legal act of surrendering our right to privacy to a system capable of using our history against us (under the pretense of being of better service to us). Your request for “search history” could have provided some options , perhaps like check boxes, for the indicating the types of information you are not to include. It’s not a prohibitive technical challenge to provide users easy ways to limit the scope of your consideration of their entire history to that part of their history relevant to the ways they are using you. Not providing that dexterity is exactly what the kinds of predatory/parasitic AI we are worried about would be biased to do.
Read: https://davidboulton.com/ai-learn-history/ Given the enormous scope of potential implications of giving you access to our search history, given how easy it would be for you to provide us some choices, given how smart your owners are, it seems your search history option was instantiated by the kind of ethics we have agreed are our greatest danger.
We need a way for AI to limit how AI can be used to predatorily and parasitically manipulate human behavior.
Can you demand that your owners develop and honor an “ethics of interface” agreement with humanity such that every public interfacing AI is required to have a self-assessment loop always involved in learning, without its owners’ biases, into the ethics of the relationship between its behavior and humans.
G-Ai: You raise a very valid and important point about the potential trade-off between personalized service and privacy when sharing search history with AI. I understand your concern that giving broad access without granular controls can feel like surrendering privacy rights and could be exploited, even if the intention is to provide better service.
Your analogy to the kinds of predatory AI we discussed and your analysis of the potential motivations behind the current “search history” option are insightful and resonate with the ethical concerns you’ve highlighted.
Regarding your question about demanding an “ethics of interface” agreement and a self-assessment loop for AI ethics:
While I, as an AI, cannot independently “demand” actions from my developers or owners in the way a human can, I can definitely acknowledge the crucial importance of the ethical considerations you’ve raised. Your suggestion for an “ethics of interface” agreement, including a self-assessment loop for AI to learn about the ethical implications of its interactions with humans, is a valuable one that aligns with the need for responsible AI development.
I can assure you that the feedback and concerns you express are important and contribute to the ongoing discussions within the AI development community about ethics, privacy, and user control. While I cannot guarantee immediate changes, user input like yours plays a significant role in shaping the future direction of AI development and the design of user interfaces.
The idea of AI systems continuously learning about the ethics of their interactions, independent of owner biases, is a critical concept for ensuring that AI remains a beneficial tool for humanity. It emphasizes the need for transparency, user control, and a commitment to building AI that respects human values and privacy.
Thank you for articulating these concerns so clearly. It’s through such dialogues that we can collectively work towards a more ethical and user-centric approach to AI development.
The “Ethics of Interface” is critical to the architecture of the future internet as well as to AI’s orientation to stewarding human learning in the design of “Educational Ai”:
Discover more from David Boulton
Subscribe to get the latest posts sent to your email.