Skip to content
FaiscaI

Risks & Ethics

5 Downsides of AI (That Few Talk About)

AI isn't all benefit. Job displacement, bias, deepfakes, energy consumption, and power concentration are real risks. See 5 downsides you need to know.

3 min read FaiscaI Editorial
Binary code flowing in dark background representing digital risks

The 5 downsides of Artificial Intelligence that most impact ordinary people in 2026 are: job displacement, disinformation and deepfakes, algorithmic bias, massive energy consumption, and power concentration in a few companies. All are real problems with no easy solution.

1. Job displacement (not fake news)

The most-discussed fear — and the most real. An IMF study estimates 40% of global jobs will be affected by generative AI.

What’s already happened:

  • Junior illustrators and designers: market shrunk 30% in 2 years
  • Blog writers: per-article rates halved
  • Junior programmers: hiring plummeted; companies prefer senior + AI
  • Customer service: thousands of call center jobs replaced

What won’t be replaced soon: manual work (plumber, hairdresser), trust professions (judge, therapist), high-level creative.

2. Industrial-scale disinformation (deepfakes)

In 2026, it’s practically impossible to look at a video and be sure it’s real. Voices, faces, gestures — all replicable with free apps.

Concrete impacts:

  • Financial scams with cloned voice of relatives: losses tripled in 2 years
  • Fake news with “videos” of politicians
  • Extortion with fake intimate images — victims are mostly teenagers

AI detection tools exist but always trail the generators.

3. Algorithmic bias (AI discriminates)

AIs learn from data. If data has bias, AI reproduces it — and amplifies it.

Documented cases:

  • Amazon’s hiring system rejected women’s résumés because it learned from a male-dominated history
  • US judicial systems recommend longer sentences for Black defendants in risk-scoring
  • Medical AIs misdiagnose darker skin because trained on lighter skin
  • Image models generate CEOs as white males and domestic workers as women of color

It’s not malice — it’s blind math repeating the past.

4. Brutal energy and water consumption

Every ChatGPT query uses energy. Scale that to billions:

  • Training GPT-4 consumed energy equivalent to 500 US homes for 1 year
  • AI data centers project to consume 8-10% of US electricity by 2030
  • Every 100 ChatGPT queries “drink” about 500ml of water for cooling
  • AI-sector CO2 emissions grew 48% in 5 years just at Microsoft

The International Energy Agency warns that without clean energy, AI will worsen climate change.

5. Power concentration in 5 companies

Virtually every AI you use comes from one of these: OpenAI, Google, Microsoft, Anthropic, or Meta. And training models costs $100 million+ — nobody else can compete.

Consequences:

  • They define what your AI can and can’t say
  • They can cut off your access without notice
  • They concentrate the economic fruits of the revolution
  • Strategic dependence for countries on foreign infrastructure

Open source movements (Hugging Face) and national AI initiatives try to break this monopoly, but they’re far behind.

Is it worth using AI anyway?

Yes, and not using it is worse. Those who understand the risks and use with awareness come out ahead. Those who pretend AI doesn’t exist will get run over — at work, by scams, in their kids’ education.

To start the right way, see:

Knowing the risks is like knowing car risks: you don’t stop driving, but you wear a seatbelt.

Frequently asked questions

Could AI ever be banned?

Specific applications already are. The EU banned mass facial recognition and social scoring by AI in the AI Act. But banning AI itself is like banning electricity — impossible in practice.

Is my data being used without my knowledge?

Probably. Most models were trained on public web content without explicit consent. Regulations are coming — the FTC and EU have issued guidance.

How do I protect myself from AI risks?

Three things: (1) don't share sensitive data in free AI, (2) doubt videos and audios — deepfakes are too good, (3) learn to use AI yourself instead of being outrun by those who do.