By:
Dr. Gleb Tsipursky
Ghana is investing in the skills needed for an artificial-intelligence economy. A government and UNESCO programme is preparing 240 civil servants to train colleagues in AI literacy, while Ghana Digital Centres and the Youth Employment Agency plan to train more than 2,000 young people in areas including AI development, data analysis and cybersecurity,
These
programmes address a real need. Yet Ghana should resist measuring success
mainly through enrolment, course completion and certificates. Those numbers
show that training occurred. They do not show whether people can use AI
responsibly in the workplace.
Every
publicly supported AI programme should end with workflow proof: a supervised
project in which a participant uses an approved tool on a real task, checks the
output, documents the risks and measures the result.
For
a civil servant, the task might involve summarising public comments, drafting a
routine briefing or organising service requests. For a small business, it could
involve inventory forecasting, customer communication or bookkeeping. For a
health worker, it might involve structuring non-sensitive information while
following medical-records and data-governance rules.
The
project should answer four questions.
First,
did the tool improve the whole workflow? Measure net time saved after
fact-checking, correction and approval. Faster drafting followed by extensive
repair creates activity rather than productivity.
Second,
did quality improve? Participants should compare error rates, completeness,
clarity and outcomes with the previous method. A confident answer deserves no
credit when it sends a citizen to the wrong office or gives a manager an
unreliable forecast.
Third,
did the user apply human judgment? The participant should explain what they
accepted, what they rejected and why. This matters because AI literacy includes
recognising uncertainty, protecting confidential information and knowing when
to escalate a decision.
Fourth,
can the organisation repeat the result safely? A useful project should produce
a short workflow guide naming the approved tool, permitted data, review steps,
accountable person and conditions that require a human-only process.
This
model would strengthen Ghana’s National AI Strategy, which seeks responsible,
human-centred adoption aligned with Ghanaian values. It would also make
training more valuable to employers. A certificate tells a hiring manager that
someone attended a programme. A documented workplace project shows that the person
can deliver an outcome under realistic constraints.
Training
providers should build partnerships with ministries, hospitals, banks, telecom
companies, farms and small enterprises to supply suitable projects.
Participants need supervised access rather than unrestricted experimentation
with sensitive data. Employers can contribute recurring tasks and review the
results. Government can aggregate anonymised evidence about which uses save
time, improve service and create risk.
The
same evidence should guide funding. Programmes that produce measurable workflow
improvements and employment transitions should scale. Programmes that generate
certificates without demonstrated capability should change. Public reporting
can include the number of completed projects, the share adopted by employers,
error reductions, jobs or promotions secured and incidents caught through human
review.
The European Union’s new AI transparency obligations, which began applying on August 2,
offer a timely reminder that disclosure alone cannot guarantee responsible use.
Ghana should go further inside the workplace. People need to understand when AI
participates in a task, who reviews it and what evidence supports the decision.
Ghana’s
advantage will not come from racing to issue the most AI certificates. It will
come from developing workers who can combine technology with context,
professional judgment and accountability.
The
practical standard is simple: no AI training programme should count a
participant as ready until that person can improve one real workflow, explain
the safeguards and show the evidence.
Training
should also include a short incident exercise. Participants should receive an
output containing a subtle factual error, privacy risk or unfair assumption and
show how they would identify, report and correct it. That test would reveal
whether a learner can protect the organisation when the tool sounds persuasive
but gets the answer wrong.
About
the writer: Dr.
Gleb Tsipursky, a behavioral scientist and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026).
