Chapter 22
Ten myths about AI, and what is true instead
The chapter in a minute
- These 10 are claims you'll hear, at parties and in the news, and most stretch something partly true too far.
- Each one has a calmer 'true instead' you can say in a sentence, and a chapter of this book that explains it.
- When you hear a new claim, ask who's saying it and why, and check the part that matters before you pass it on.
Try it now
Myth or true?
0 of 4 steps done
Step 1 of 4
Know it. Mark each claim myth or true, from what you know about these tools now.
Try this
Try this, from the book
If you'd like to see what your assistant remembers about you, have a look in its settings this week. It takes a couple of minutes, and you can change those notes or switch memory off.
Tools and settings, as of a date
The book points to the ‘Myths and facts’ section of the companion page. Here it is in full, with the date each detail was checked. Open it on the companion page
Chapter 22 points here, with the survey figures and memory features. Survey figures accessed 24 October 2026; the Pew UK row and the memory pages read again on 2 November 2026.
What people expect about jobs
- UK: 61% expect fewer jobs over 20 years (spring 2026). US: 71% (June 2026). US 64% and experts 39% (2024). "In 34 of 37 countries" people tend to expect fewer jobs. Sources: Pew Research Center, "Globally, More People Expect AI to Cause Job Loss Than Growth (opens in a new tab)", 17 September 2026 (UK row confirmed in the topline PDF (opens in a new tab) on 2 November 2026: 7% more jobs, 61% fewer, 9% no real difference, 24% not sure), and Pew, "How the U.S. Public and AI Experts View Artificial Intelligence (opens in a new tab)", 3 April 2025.
- UK 2024 tracker (fieldwork 2024, an older wave): 40% foresaw a negative effect on their family's job opportunities. Public attitudes to data and AI, wave 4 (opens in a new tab)
Risks and reasons for not using
- UK perceived risks of AI: inaccuracy 74%, job loss 63%, no concerns 3%. Reasons UK non-users give: 32% do not know enough; 44% prefer without AI; 33% privacy; 31% accuracy. DSIT Public Engagement Survey 2025/2026 (opens in a new tab), published 16 July 2026, push-to-web survey of UK adults aged 16 and over, November 2025 to March 2026. Next release expected 2027.
- US reasons for not using chatbots: not interested 83%; accuracy 76%; not knowing how 55%; others judging them 14%. Pew, "Why Don't Some People Use Chatbots? (opens in a new tab)", 9 October 2026, from a survey of 5,119 U.S. adults, 17 to 23 February 2026.
- UK 2024 tracker: one-word feelings about AI, with "robot" mentions falling, and seven in ten knowing little or nothing about how AI is trained.
Search answers, trust and use
- 50% trust AI answers in search; about one in three click the links regularly; UK 40% trust, 32% distrust. Reuters Institute, Generative AI and News Report 2025 (link under "Everyday AI"). The 2026 edition, "From standalone tools to information infrastructure (opens in a new tab)", came out on 29 October 2026. It covers nine countries including the UK but does not repeat the search-answer trust figures in the same form, so the 2025 figures stand here. It does report that UK weekly use of AI tools, at 42%, was the lowest of the nine countries.
- Use: US 49% use chatbots, 42% to search. UK 59% used generative AI in the previous three months. Pew, "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact (opens in a new tab)", 17 June 2026, and DSIT, above.
- What comes to mind when Americans hear "AI": chatbots 29%; robots and science fiction 8%. Same Pew report.
- US views on the pace: 63% say too quickly; 2% too slowly. Same Pew report.
Does a chat assistant have feelings?
A 2023 study of whether people think a chat assistant may have experiences: 67% of 300 US participants. Colombatto and Fleming, "Folk psychological attributions of consciousness to large language models". Small and dated.
Memory features, by provider
As of 2 November 2026. These change often.
| OpenAI (ChatGPT) | Anthropic (Claude) | |
|---|---|---|
| What it remembers | Relevant preferences and details from your chats; depending on plan and region, other sources such as files and connected apps. "Memory does not retain every detail from every conversation." | Individual topics as you chat, such as your role, projects and preferences |
| On or off to start | Controls vary by plan, region, platform and workspace. | On by default for the Free, Pro, Max and Ultra plans; on Team and Enterprise plans the organisation controls it and it is off for each member until they turn it on |
| Sensitive topics | Sensitive information you share can end up in memory; the page suggests an unpersonalised temporary chat when you don't want memories made | By default it does not store sensitive topics such as health, religion or politics unless you choose to; government ID numbers, criminal history, financial account numbers and immigration status are never saved |
| Pausing, resetting | You can ask what it remembers, and review, correct or delete remembered items | You can pause memory (kept but not used or added to), reset it (which permanently deletes it), or turn it off for a single chat |
| Temporary or incognito | See "Settings" | An incognito chat is not saved to memory or history |
| Turning memory off | Does not delete your past chats | Pausing keeps what is already remembered; only a reset deletes it |
| Deleting a chat | Does not necessarily delete a memory made from it | Memory made from a deleted chat is not removed; you can delete single memories yourself |
Pages: OpenAI, Memory in ChatGPT (opens in a new tab) and Anthropic, chat search and memory (opens in a new tab).
The 2024 training data study
Longpre and colleagues, "A Pretrainer's Guide to Training Data: Measuring the Effects of Data Age, Domain Coverage, Quality, and Toxicity (opens in a new tab)", NAACL 2024. The team trained 28 small models from scratch, each on differently prepared text, to see what the choice of training text does. In short: models do worse on text newer than their training data, and later tuning doesn't fully fix it; filtering out low-quality text helped on most tasks even though it left less text to learn from; filters trade one thing against another, and a filter that improved quality also raised harmful output; and a mix of well-edited, varied sources such as books and web pages worked best. The models were much smaller than today's assistants, so read it as evidence that what goes in matters, not as a measure of any product. Read on 2 November 2026.
From the book
Sounding sure is not evidence. A confident tone tells you how the answer was written. Whether it's true is something you find out by checking.
Sources
- Incident 1028: OpenAI's Operator Agent Reportedly Executed Unauthorized $31.43 Transaction Despite Safety Protocol (opens in a new tab) · AI Incident Database, 2025
- Context windows (opens in a new tab) · Anthropic, 2026
- Enable and use web search (opens in a new tab) · Anthropic, 2026
- Glossary (opens in a new tab) · Anthropic, 2026
- How long do you store my data? (opens in a new tab) · Anthropic, 2026
- Use Claude's chat search and memory to build on previous context (opens in a new tab) · Anthropic, 2026
- Models overview (opens in a new tab) · Anthropic, 2026
- 'AI' as a Marketing Term: A Quantitative Review of Usage in UK Advertising (opens in a new tab) · Advertising Standards Authority, 2024
- Careful adoption of agentic AI services (opens in a new tab) · Australian Signals Directorate's Australian Cyber Security Centre and Cybersecurity and Infrastructure Security Agency and National Security Agency and Canadian Centre for Cyber Security and National Cyber Security Centre (New Zealand) and National Cyber Security Centre (UK), 2026
- Language Models are Few-Shot Learners (opens in a new tab) · Brown, Tom B. and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom and Child, Rewon and Ramesh, Aditya and Ziegler, Daniel M. and Wu, Jeffrey and Winter, Clemens and Hesse, Christopher and Chen, Mark and Sigler, Eric and Litwin, Mateusz and Gray, Scott and Chess, Benjamin and Clark, Jack and Berner, Christopher and McCandlish, Sam and Radford, Alec and Sutskever, Ilya and Amodei, Dario, 2020
- Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence (opens in a new tab) · Brynjolfsson, Erik and Chandar, Bharat and Chen, Ruyu, 2026
- Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) · Department for Science, Innovation and Technology, 2024
- DSIT Public Engagement Survey 2025/2026 (opens in a new tab) · Department for Science, Innovation and Technology, 2026
- Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) (opens in a new tab) · European Parliament and Council of the European Union, 2024
- Individuals - use of generative AI tools (opens in a new tab) · Eurostat, 2026
- News Integrity in AI Assistants: An international PSM study (opens in a new tab) · Fletcher, James and Verckist, Dorien, 2025
- The future of employment: How susceptible are jobs to computerisation? (opens in a new tab) · Frey, Carl Benedikt and Osborne, Michael A., 2017
- Artificial Intelligence Playbook for the UK Government (opens in a new tab) · Government Digital Service, 2025
- Generative AI and Jobs: A Refined Global Index of Occupational Exposure (opens in a new tab) · Gmyrek, Pawe\l and Berg, Janine and Kami\'nski, Karol and Konopczy\'nski, Filip and \Ladna, Agnieszka and Nafradi, Balint and Ros\laniec, Konrad and Troszy\'nski, Marek, 2025
- Gemini Apps Privacy Hub (opens in a new tab) · Google, 2026
- Americans and AI 2026: Chatbots, Smart Devices and Views on Impact (opens in a new tab) · Gottfried, Jeffrey and Bishop, William and Anderson, Monica and Faverio, Michelle and Park, Eugenie and McClain, Colleen, 2026
- The 2026 AI Index Report (opens in a new tab) · Stanford Institute for Human-Centered Artificial Intelligence, 2026
- Survey of Hallucination in Natural Language Generation (opens in a new tab) · Ji, Ziwei and Lee, Nayeon and Frieske, Rita and Yu, Tiezheng and Su, Dan and Xu, Yan and Ishii, Etsuko and Bang, Ye Jin and Madotto, Andrea and Fung, Pascale, 2023
- Evaluating large language models for accuracy incentivizes hallucinations (opens in a new tab) · Kalai, Adam Tauman and Nachum, Ofir and Vempala, Santosh S. and Zhang, Edwin, 2026
- A Pretrainer's Guide to Training Data: Measuring the Effects of Data Age, Domain Coverage, Quality, & Toxicity (opens in a new tab) · Longpre, Shayne and Yauney, Gregory and Reif, Emily and Lee, Katherine and Roberts, Adam and Zoph, Barret and Zhou, Denny and Wei, Jason and Robinson, Kevin and Mimno, David and Ippolito, Daphne, 2024
- Mata v. Avianca, Inc., No. 22-cv-1461 (PKC): Opinion and Order on Sanctions (opens in a new tab) · United States District Court for the Southern District of New York, 2023
- What Do NLP Researchers Believe? Results of the NLP Community Metasurvey (opens in a new tab) · Michael, Julian and Holtzman, Ari and Parrish, Alicia and Mueller, Aaron and Wang, Alex and Chen, Angelica and Madaan, Divyam and Nangia, Nikita and Pang, Richard Yuanzhe and Phang, Jason and Bowman, Samuel R., 2023
- The debate over understanding in AI's large language models (opens in a new tab) · Mitchell, Melanie and Krakauer, David C., 2023
- ChatGPT and large language models: what's the risk? (opens in a new tab) · Chismon, Dave and Paul J, 2023
- Dissecting racial bias in an algorithm used to manage the health of populations (opens in a new tab) · Obermeyer, Ziad and Powers, Brian and Vogeli, Christine and Mullainathan, Sendhil, 2019
- Introducing ChatGPT search (opens in a new tab) · OpenAI, 2024
- Deleting and archiving chats in ChatGPT (opens in a new tab) · OpenAI, 2026
- How ChatGPT and our foundation models are developed (opens in a new tab) · OpenAI, 2026
- Memory in ChatGPT (opens in a new tab) · OpenAI, 2026
- Globally, More People Expect AI to Cause Job Loss Than Growth (opens in a new tab) · Silver, Laura and Wike, Richard and Fagan, Moira and Lippert, Jordan, 2026
- Generative AI and News Report 2025: How People Think About AI's Role in Journalism and Society (opens in a new tab) · Simon, Felix M. and Nielsen, Rasmus Kleis and Fletcher, Richard, 2025
- ELIZA: A Computer Program for the Study of Natural Language Communication Between Man and Machine (opens in a new tab) · Weizenbaum, Joseph, 1966
- OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments (opens in a new tab) · Xie, Tianbao and Zhang, Danyang and Chen, Jixuan and Li, Xiaochuan and Zhao, Siheng and Cao, Ruisheng and Hua, Toh Jing and Cheng, Zhoujun and Shin, Dongchan and Lei, Fangyu and Liu, Yitao and Xu, Yiheng and Zhou, Shuyan and Savarese, Silvio and Xiong, Caiming and Zhong, Victor and Yu, Tao, 2024
- An Illusion of Progress? Assessing the Current State of Web Agents (opens in a new tab) · Xue, Tianci and Qi, Weijian and Shi, Tianneng and Song, Chan Hee and Gou, Boyu and Song, Dawn and Sun, Huan and Su, Yu, 2025
- WebArena: A Realistic Web Environment for Building Autonomous Agents (opens in a new tab) · Zhou, Shuyan and Xu, Frank F. and Zhu, Hao and Zhou, Xuhui and Lo, Robert and Sridhar, Abishek and Cheng, Xianyi and Ou, Tianyue and Bisk, Yonatan and Fried, Daniel and Alon, Uri and Neubig, Graham, 2023