logo
Don't believe everything AI tells you: A cautionary tale for academia

Don't believe everything AI tells you: A cautionary tale for academia

Mail & Guardian5 days ago
Artificial intelligence can be a powerful ally but only if we cultivate the skills and habits that affirm our commitment to truth, discernment and verification. Graphic: John McCan/M&G
I recently sat in a departmental colloquium where students were defending their research proposals before a panel of academics. Anyone who has gone through this exercise will attest the process of defending your master's or PhD proposal is, at best, a daunting and nerve-racking experience.
The task is simple in theory but difficult in practice. The panel is seeking the student to prove their proficiency in conducting the research and clearly showing the gap their proposed study addresses. All this, within 10 to 15 minutes, to an audience in the room (mostly online nowadays) but also an audience that is referred to as the theory, policy and practitioner press. In the corner of the student (hopefully) are the watchful eyes and muted voices of their supervisor or supervision team, who themselves stand on trial before their academic peers. The result is a delicate dance, where the spoken word must align seamlessly with the written proposal.
As one student delivered their presentation, my attention was caught by their mention of an article allegedly authored by me, published in the Journal of Business Ethics. A quick glance at their supporting documents confirmed my worst fear. I have never published a paper in that journal. Further to this, I don't even research or write in the field of business ethics.
So, what had happened?
The student had fallen victim to what is now widely known as an AI hallucination. In simple terms, they had placed their trust in the output of an artificial intelligence tool, which generated what looked like credible information about their topic and about me but which was fabricated.
For the student, the AI-generated information seemed real. It said all the 'right' things and cited the kind of references a proposal defence panel would expect to hear and see. Yet, the result was false, misleading and nonsensical. What was missing was a critical process of verification needed long before the student could even be deemed to be ready to take part in this proposal defence.
What we saw here was a double-layered false confidence. First, the false confidence of the AI itself. This came in the form of confidently making connections based on user prompts, some factual, others wholly fictional. Second, the false confidence of the human user through presenting AI hallucinations as fact, without adequate scrutiny, driven perhaps by the desire to impress a panel at all costs.
What happened to the student?
I choose to reflect on that last, because what happened to us as supervisors was equally instructive and worth reflecting upon. The experience (including the imaginary Journal of Business Ethics paper) became, for me, what sociologist Charles Horton Cooley called a 'looking glass self'. I began to see aspects of myself and my supervision practice through the mirror held up by the student's mistake.
I prefer to describe what the student did as a mistake, rather than a punishable offence or as one leading survey in the United Kingdom called it, a violation of academic integrity. This incident sparked months of reflection for me.
In a sobering way, I realised that my own experience with AI was not so different from the student's. Like our students, we supervisors are also searching for timely information to meet pressing demands. Like our students, we too struggle under the weight of information overload, turning to tools like AI to help us navigate the maze. And, like our students, we must also develop and exercise a critical eye in the face of what may appear to be technological progress.
How did we respond as supervisors?
For starters, given the growing popularity of AI among our students, some of us as supervisors felt the need to use such technology ourselves, to stay abreast of changes in the academic and professional landscape. It meant moving out of our comfort zones into spaces of discomfort, just to keep pace with what is happening.
Some supervisors were quick to praise the functionality AI offers. For instance, using an AI tool to analyse large amounts of data in a short space of time was seen as a significant benefit. Others highlighted how AI could help students develop their writing and critical thinking skills provided that students' own voices remained central to the work, rather than being drowned out by the machine-generated content.
We are truly living at the height of a technological moral panic, a time when our ability to exercise our executive functioning skills is being eroded precisely when we need them the most. It is a period in which voices of falsehood are legion, spreading at the mere click of a button, often without verification or reflection. Yet, this is also the very moment when we must be most vigilant and rise to the task of cultivating the skills and habits that affirm our commitment to truth, discernment and verification.
Through the experience of watching students present their research proposals, we came to realise that our struggles are, in fact, the same; they just take different forms. As supervisors in our department, we embarked on a month-long dialogue with our students, acknowledging and praising the benefits of AI while also cautioning them about the dangers of AI hallucinations. Our hope is that this process proves beneficial for everyone involved. This benefit is anchored in helping students, supervisors, the university and ultimately society at large to achieve success rooted in both innovation and integrity.
AI can be a powerful ally but only if we, both students and supervisors, treat its outputs as a starting point for inquiry, not the final word.
Professor Willie Chinyamurindi is in the Department of Applied Management, Administration and Ethical Leadership at the University of Fort Hare. He writes in his personal capacity.
Orange background

Try Our AI Features

Explore what Daily8 AI can do for you:

Comments

No comments yet...

Related Articles

Elon Musk's Starlink satellite internet suffers global outage
Elon Musk's Starlink satellite internet suffers global outage

The South African

time6 hours ago

  • The South African

Elon Musk's Starlink satellite internet suffers global outage

SpaceX chief Elon Musk's Starlink satellite internet service experienced an hours-long global network outage late on Thursday, which executives attributed to a key software issue. The service interruption was announced on X at about 22:00 (SA time) on Thursday by Starlink's official handle. Users in the United States and Europe began reporting problems with the service an hour before the Starlink announcement on Downdetector, a website that tracks issues in internet-based services. 'Service will be restored shortly,' South Africa-born Musk posted on X, apologising for the outage. 'SpaceX will remedy root cause to ensure it doesn't happen again.' The tech billionaire later reposted a statement from Starlink Vice President of Engineering Michael Nicolls, who said the disruption was due to the 'failure of key internal software services that operate the core network.' Nicolls also said the Starlink network had 'mostly recovered' from the outage, which 'lasted approximately 2.5 hours.' About two hours later, Starlink posted that the issue was resolved and that service was restored. User reports on Downdetector began after 21:00 (SA time), peaking at 21:34 (SA time) and tapering out at 05:00 (SA time). Starlink, a subsidiary of Musk's space rocket venture SpaceX, has deployed more than 6 000 low-orbit satellites to provide high-speed internet to isolated and poorly connected areas. Starlink currently leads the satellite internet race, with European competitor Eutelsat – which is backed by France and the United Kingdom – lagging behind with 600 satellites. Let us know by leaving a comment below, or send a WhatsApp to 060 011 021 1 Subscribe to The South African website's newsletters and follow us on WhatsApp, Facebook, X and Bluesky for the latest news. By Garrin Lambley © Agence France-Presse

Discover how AI-Powered analytics are revolutionising retail Forex Trading in 2025
Discover how AI-Powered analytics are revolutionising retail Forex Trading in 2025

Mail & Guardian

time13 hours ago

  • Mail & Guardian

Discover how AI-Powered analytics are revolutionising retail Forex Trading in 2025

The foreign-exchange market has always rewarded traders who can process information faster than the competition. In 2025, that advantage increasingly belongs to those who deploy artificial-intelligence-driven analytics. South African retail traders, once limited to basic charting packages and delayed economic calendars, now enjoy real-time pattern detection, sentiment scoring, and predictive order-flow models piped straight into their trading dashboards. What was once the preserve of hedge-fund quants is being delivered through cloud-first platforms that cost less per month than a Saturday braai with friends. Platforms such as The traditional technical-analysis toolkit still matters, but AI augments it in ways that speak directly to the needs of a retail audience. Deep-learning models trained on decades of market microstructure spot hidden liquidity pockets long before conventional indicators flash. Natural-language-processing engines scan every Monetary Policy Committee briefing out of Pretoria and every Treasury press release, tagging each sentence with a probability that it will move USD ZAR by more than 0.5 %. These probabilities appear as colour-coded prompts beside a trader's chart, trimming hours from the research cycle and reducing the urge to trade on gut feeling alone. South Africa boasts the largest, most liquid Predictive order-book heatmaps draw on Level-II data from Johannesburg-hosted servers, showing where institutional bots are clustering liquidity. Sentiment fusion models combine Twitter, YouTube, and local news feeds to score public mood about the rand, gold prices, and the broader BRICS narrative. Reinforcement-learning trade managers adjust stop-loss and take-profit levels dynamically, factoring in S&P 500 futures, Bitcoin momentum, and commodity index volatility. Voice-activated assistants let traders ask, 'What is the probability of a USD ZAR rally if Brent crude jumps three per cent?' and receive an instant, data-backed answer. These tools lower the knowledge barrier for beginners while giving veterans sharper edges for prop-firm assessments. AI is not merely about finding entries; it transforms risk control, too. Value-at-risk models that once refreshed daily now recalculate every minute, pulling live volatility surfaces from interbank venues. When sudden rand weakness lifts the three-month implied volatility curve, the smart system can automatically cut leverage on open positions or recommend a counter-correlated hedge in AUD USD. For South Africans trading from their phones between meetings, this auto-adjustment is a lifeline, preventing small mistakes from compounding into blown accounts. Moreover, AI exposes hidden correlations that local traders may overlook. In 2024, an unscheduled power-station outage in Mpumalanga coincided with a spike in inland coal prices, which in turn pressured mining-linked equities and dragged the rand lower. An AI engine that maps energy supply data to currency performance could have pre-emptively reduced exposure hours before the sell-off became obvious on charts. Looking toward the second half of 2025, three trends are set to intensify. First, edge computing under the POPIA framework will keep more personal data within South Africa's borders while still feeding anonymised trade statistics to global AI clouds, balancing privacy and performance. Second, generative AI agents will start running full back-tests and generating narrative reports that explain their logic in plain English, helping traders satisfy both compliance and client-reporting requirements without manual spreadsheets. Third, broker-agnostic neural APIs will allow savvy coders to plug AI signals directly into MT5 or cTrader, opening the door to semi-autonomous portfolios that self-rebalance based on volatility clustering instead of fixed time intervals. For retail traders, the message is clear: mastering the basics of forex remains essential, but harnessing AI analytics is quickly becoming the decisive differentiator. Those who embrace the new tools stand to capture moves earlier, size positions more intelligently, and protect capital during South Africa's trademark bouts of volatility. Traders who ignore the shift risk being outpaced by algorithms that never sleep. In a market where milliseconds matter and global flows can spin the rand from calm to chaos before dawn breaks over Table Mountain, AI-powered analytics offer a competitive edge that is both accessible and increasingly indispensable. Whether you are a newcomer placing your first micro-lot or an experienced day-trader aiming for consistent withdrawals in 2025, integrating intelligent datafeeds into your strategy could be the upgrade that turns potential into real, measurable performance.

Hackers target Treasury, servers isolated as team assesses malware
Hackers target Treasury, servers isolated as team assesses malware

The Herald

timea day ago

  • The Herald

Hackers target Treasury, servers isolated as team assesses malware

The National Treasury says it identified malware on its infrastructure reporting and monitoring (IRM) website on Tuesday afternoon. The IRM servers were isolated for its staff to assess the magnitude of the compromise and to ensure the security of its systems. In light of Microsoft's recent warning of attacks on software globally, the Treasury said it has contacted the tech giant to check for and address potential vulnerabilities in its information and communication technology (ICT) environment. Despite this, the Treasury said its systems and websites continue to operate normally without any disruption. 'The National Treasury's ICT department processes more than 200,000 e-mails each day and facilitates more than 400,000 user connections through their websites daily,' it said. 'On average, the ICT team successfully detects and blocks about 5,800 security threats directed at Treasury systems every day, showcasing the department's commitment to maintaining a secure digital environment. These threats encompass a range of malicious activities, including phishing attempts, malware infections and spam attacks.' TimesLIVE

DOWNLOAD THE APP

Get Started Now: Download the App

Ready to dive into a world of global content with local flavor? Download Daily8 app today from your preferred app store and start exploring.
app-storeplay-store