AI Myths vs Facts: 12 Common Misconceptions Checked

By the WahLiao desk · Last verified 3 October 2026

A lot of what people “know” about Super Intelligence, the kind of AI behind ChatGPT, Claude and Gemini, is only half right. Here’s what we found. It doesn’t understand things the way a person does, it isn’t conscious, it doesn’t search the internet for every answer, and it isn’t right just because it sounds sure. But it’s also not “just autocomplete”, it isn’t banned in Singapore’s schools, and you don’t need to be technical to use it. The free plans cost S$0; paid plans start at S$6.98 a month for Google AI Plus. Below are 12 common myths, each one answered with a fact and the source we checked it against.

If you remember one thing, make it this: judge Super Intelligence by what it actually does on your task, then double-check anything that matters. As far as we can tell, neither the hype nor the eye-rolling holds up once you look at the evidence. And if we’ve got something wrong, tell us and we’ll fix it.

AI myths and facts: quick facts

How it works	Predicts text from patterns learned in training; can plan ahead internally
Conscious?	No current system is, according to a 2023 review by 19 researchers
Confident mistakes	OpenAI says ChatGPT may sound confident even when it’s wrong
AI detectors	Unreliable; in one study, detectors on average flagged 61% of non-native English essays as AI
Jobs	WEF forecast: 170 million jobs created, 92 million displaced by 2030
Schools	No ban; MOE uses its own AI tools in SLS under teacher supervision
Cost to start	S$0; paid plans from S$6.98 a month

The WahLiao Verdict

Biggest myth	“It sounds sure, so it must be right.”
Most expensive myth	“Paid is always better.” Try the free plan first.
Most unfair myth	“The detector says AI, so the student cheated.”
Most dangerous myth	“I can always spot a fake picture.”
Healthiest stance	A useful tool. Not a person, and not an oracle.

12 AI myths vs facts at a glance

Myth	Fact
It understands like a person	It learns patterns from training data; it has no lived experience and can trip up in very un-human ways
It’s conscious	Researchers found no current system meets the indicators
It’s just autocomplete, so useless	It predicts one word at a time, but can plan several steps ahead
It searches the internet every time	ChatGPT searches only when it judges it useful, or when you ask it to
It’s right if it sounds confident	Sounding sure isn’t the same as being right; training rewards guessing
It will take all jobs next year	The WEF expects churn to 2030 and a net gain of 78 million jobs
AI detectors reliably catch AI writing	OpenAI withdrew its own detector for low accuracy
Singapore bans ChatGPT in schools	No ban; MOE brings in its own AI tools by age
Paid plans are always better	ChatGPT Free includes search, uploads and images; paid adds higher limits and speed
It’s private by default	Consumer chats may be used for training unless you opt out
You need to be technical	You type in plain language; free, self-paced courses need no coding
AI images are easy to spot	In a PNAS study, people scored 48%, no better than a coin toss

Myths about how it thinks

Myth 1: “It understands like a person”

OpenAI describes ChatGPT as designed to give useful responses based on patterns in the data it was trained on. That’s not how you or I understand things. It has no body, no lived experience and no stake in being right, so it can handle a hard question nicely and then stumble on an easy one. If you’d like the plain-language version of what goes on under the hood, our explainer on how Super Intelligence works walks through training and prediction.

Myth 2: “It’s conscious”

In 2023, 19 researchers from philosophy, neuroscience and computer science checked leading AI systems against indicators drawn from scientific theories of consciousness. Their conclusion: no current AI systems are conscious. They also wrote that they saw no obvious technical barriers to building systems that meet those indicators, so in their view this is an open question for the future rather than one settled for good. The separate idea of a far-beyond-human “superintelligence” has its own page: AGI and superintelligence explained.

Myth 3: “It’s just autocomplete, so it’s useless”

This one is half true. Models are trained to produce one word (strictly, one token) at a time. But when Anthropic’s interpretability team looked inside Claude in March 2025, they reported that it plans ahead: when writing rhyming verse, it picked candidate rhyming words before writing the line, and it answered two-step questions by linking facts in sequence. So “autocomplete” describes what comes out, not everything happening inside. For tips on getting good results, see using Super Intelligence well.

Myths about accuracy and search

Myth 4: “It searches the internet every time”

It doesn’t. OpenAI’s help centre says ChatGPT may search the web on its own when your question would benefit from current information; otherwise it answers from what it learned in training. You can make it search by choosing Search from the tools menu. A handy habit: if a reply has no links to sources, assume it didn’t search, and treat any prices, opening hours or rules in it with a healthy pinch of salt.

Myth 5: “It’s always right if it sounds confident”

OpenAI’s own help centre says ChatGPT may sound confident even when it’s wrong, and it lists made-up quotes, studies and citations among the known errors. A September 2025 OpenAI research paper explains part of the reason: standard training and testing reward a guess over admitting uncertainty, a bit like a multiple-choice paper where a guess can score but a blank never does.

Myths about work and school

Myth 6: “It will take all jobs next year”

We couldn’t find any credible forecast that says this. The World Economic Forum’s Future of Jobs Report 2025, based on a survey of employers, expects 22% of jobs to be disrupted by 2030, with 170 million roles created and 92 million displaced, a net gain of 78 million. The same survey found 41% of employers plan to cut staff because of AI automation, while 77% plan to upskill their workers. These are forecasts, not promises, so treat them as one informed view. For what Singapore is doing about it, see Budget 2026 and your job and careers in Super Intelligence.

Myth 7: “AI detectors can reliably catch AI writing”

OpenAI launched its own AI-text classifier in 2023. By its own figures, it correctly identified only 26% of AI-written text and wrongly flagged 9% of human writing, and OpenAI withdrew it on 20 July 2023 for low accuracy. A Stanford study of seven detectors found that, on average, they misclassified 61% of TOEFL essays by non-native English writers as AI-generated, while doing almost perfectly on essays by US eighth-graders; 97% of the TOEFL essays were flagged by at least one detector. For Singapore students who write in a second language, that matters: a detector score on its own isn’t proof.

Myth 8: “Singapore bans ChatGPT in schools”

There’s no ban, and MOE hasn’t announced one. In 2023 MOE told Parliament these tools can support learning once students have mastered basic concepts and thinking skills. By 2026 the approach is graded by age: in lower primary, pupils don’t use AI directly; from Primary 4, use is limited to MOE-developed tools in the Student Learning Space (SLS), in class and under teacher supervision. National exams stay proctored and AI use in them is prohibited, and students who pass off AI work as their own face academic-dishonesty consequences. School rules can change, so check MOE’s latest guidance before you act. Parents may find children and Super Intelligence helpful.

Myths about money, privacy and skill

Myth 9: “Paid plans are always better”

Paid plans do give you more: OpenAI says ChatGPT Plus adds faster performance, expanded tools and higher limits. But the free plan already includes web search, file and image uploads and image generation, within usage limits. Plans and prices change often, so before you pay, compare free and paid plans in S$ and check whether you qualify for six months free after an eligible SkillsFuture AI course.

Myth 10: “It’s private by default”

On consumer ChatGPT plans, including Free and Plus, OpenAI says it may use your content to train its models unless you switch off “Improve the model for everyone” in Data Controls. Inputs to ChatGPT Business, Enterprise, Edu and the API aren’t used for training by default. Even with training off, your chats still go to the provider and are stored there, so keep NRIC numbers, passwords and client files out of them. More in your data and the PDPA.

Myth 11: “You need to be technical”

You talk to it in everyday language, the same way you’d message a friend about where to meet for kopi. AI Singapore’s AI for Everyone course is free, online and self-paced, and it’s pitched at a layperson, with no coding. Seniors can start with Super Intelligence for seniors; if you like a bit of structure, try a 90-day learning path.

Myths about images

Myth 12: “AI images are easy to spot”

In a study published in PNAS in 2022, 315 people asked to tell real faces from AI-generated ones scored 48%, slightly worse than chance; a second group who were given training reached only 59%. Participants also rated the synthetic faces as more trustworthy. So it’s kinder to yourself not to rely on your eye: check through a separate channel instead, as we set out in how to check for deepfake scams.

AI myths and facts: FAQ

Is AI conscious or sentient?

As far as current research goes, no. A 2023 review that checked leading models against scientific theories of consciousness found no current system that is.

Does ChatGPT search the internet for every answer?

No. It searches when it judges current information would help, or when you choose Search yourself.

Are AI detectors accurate?

Not reliably. OpenAI withdrew its own detector in 2023 for low accuracy, and a Stanford study found detectors often flag non-native English writing as AI.

Is ChatGPT banned in Singapore schools?

No. MOE brings AI in by age: none directly in lower primary, then MOE-developed tools in SLS from Primary 4 under teacher supervision. AI is prohibited in national exams.

Will AI take my job?

The forecasts we checked point to change rather than wholesale loss. The WEF expects 92 million jobs displaced and 170 million created worldwide by 2030, though forecasts like this can be wrong.