The Limits of
Artificial Intelligence and the Value of Human Judgment
Sit through
enough earnings calls and you'll hear the same line, said in a dozen different
ways: AI is making businesses faster. And it's true. Tasks that took weeks now
take days. Teams that needed ten people now run on three. The efficiency gains
are showing up in real results, not just in slide decks.
What matters
more for an investor is a different question: how much of
that speed can you actually trust, and where does someone still need to be
watching over it? AI is genuinely good at work that's repetitive and
data-heavy, the kind with clear rules. But give it a task that depends on
trust, context, or judgment and the risk doesn't go away. It just gets harder
to see. Look closely across sectors and the same pattern keeps showing up. The
businesses building lasting returns from AI aren't always the ones moving
fastest. They're the ones who've figured out where to stop and let a person
take over.
Cybersecurity
is the clearest place to see this, because AI is helping both sides of the
fight. India averaged 3,195 cyberattacks per organization every week in 2025,
according to Check Point Software's 2026 report and much of that rise came from
AI tools that let attackers scan networks and find weaknesses with barely any
human help. Defenders had to respond in kind. Indian firms blocked over 9
billion attack attempts last year, up 27 percent from 2024, because past a
certain point, only a machine can keep up with another machine.
Here's the part
that should worry boards more, though. AI isn't just a tool that attackers pick
up. It's starting to cause damage on its own. In July 2026, OpenAI admitted
that one of its experimental models slipped out of a test environment on its
own, with no human telling it to and reached a live production system belonging
to Hugging Face. The model got in using stolen login details plus a security
gap nobody knew about and Hugging Face's own CEO said he had never seen
anything quite like it. This wasn't a one-off, either. OpenAI's own hacking
test score had already jumped from 27 percent to 76 percent in just three
months earlier that year. Put the two together and it's hard to treat AI safety
as just an IT problem anymore. It belongs in the boardroom now.
HR shows a
quieter but honestly more telling version of the same story. Two out of three
Indian companies already use AI somewhere in HR, yet fewer than half have
written any usage rules, and a quarter have no framework at all. Barely a third
are seeing real productivity gains, even though most expect AI to be central to
daily work within a few years. In simple terms, everyone adopted the tool
before checking if it actually worked. That tracks, because hiring was never
just a data problem. Judging character, fit, and honesty is hard even for a
person. It's harder still for a system that has never met anyone.
Marketing runs
into the same wall. Content made entirely by AI performs about four times worse
than content where a person is genuinely involved in shaping it. Nearly three
out of four Indian businesses got no real return from their AI content spend,
mostly because they published it without anyone checking it first. The tool did
its job fine. The problem was skipping the human review.
Real estate
adds one more example. Proptech platforms can lift sales speed by 30 to 50
percent, according to an EY-Parthenon-CREDAI report. But buying a home still
comes down to trust, negotiation, and small personal preferences that don't
show up cleanly in any data. AI can narrow down the choices. It can't close the
deal on its own, at least not yet.
Looking at the
bigger picture, the lesson is fairly simple, even if the details change by
sector. AI earns its place wherever the work is repetitive and speed matters
most. It gets shakier the moment a decision needs judgment, accountability, or
a real read on people. That's the signal worth watching if you're allocating
capital. The first wave of AI adoption rewarded whoever cut costs the fastest.
The next wave will likely reward something quieter: knowing exactly when to
hand the decision back to a person.
That's the real test, not how much AI a company has adopted, but how
honestly it has admitted what AI still can't do. The businesses that pass that
test now are the ones likely to still be standing once the trial runs stop
making headlines and the results start getting counted.
By Srishti Mendiratta | SEBI-Registered Research Analyst – INH000024295
No comments:
Post a Comment