By 
By Vanessa Jacklin-Levin, Partner, and Mbali Mnyandu, Senior Associate, Bowmans
For a growing number of organisations seeking greater efficiencies and effectiveness, artificial intelligence (AI) is proving to be a brilliant tool, but no matter how good AI gets, it is not going to displace the human element any time soon.
RELATED: Global economic outlook darkens as geopolitical risks rise but AI still offers long-term hope — WEF Report
Here are six high-stakes business scenarios where human intervention should be non-negotiable for employers wanting to leverage AI while minimising the risks and potential liabilities.
Using AI to make decisions on HR and employment-related issues
 Increasingly, employers are considering using AI to drive decision-making for human resources or employment relations processes, including recruitment and hiring, retrenchment selection, performance monitoring and even dismissals.
The risk here lies not in the use of AI in itself but rather in using it blindly. Employers must have policies that clearly define the parameters within which AI tools are going to be used and must put the boundaries in place before the AI tool is switched on and not afterwards when disputes arise.
In all instances, the non-negotiable requirement is that human beings must be kept in the loop.
The organisation must designate a person able to understand, explain and defend the criteria and considerations the AI tools take into account when generating their outputs.
In the South African context particularly, every decision made with an impact on an employee must have a reasonable and justifiable basis that an employer can articulate and defend. Simply saying ‘the algorithm said so’ will not withstand scrutiny at the Commission for Conciliation, Mediation and Arbitration or the Labour Court.
Confidential information ending up on open AI platforms
 We are seeing a surge in employees using AI tools in their day-to-day work. While this can create efficiencies, it can also give rise to risks, such as the risk of employees inputting company-confidential information onto open AI platforms.
It is critical for companies to have policies in place to proactively prevent such risks and clearly spell out the consequences for employees who use AI in breach of those policies.
Generating arguments in disciplinary cases
 Another emerging trend is employees using AI to produce arguments for internal processes such as disciplinary cases. Companies then have to expend significant resources trying to debunk what seem to be sophisticated legal arguments but are in fact AI hallucinations.
Again, guardrails for this kind of practice need to be spelt out in company policies and human judgement should be applied to distinguish fact from fiction.
Clouding the waters in whistleblower complaints
 It can be problematic when employees or other parties use AI to compile whistleblower complaints. When used for this purpose, AI automatically searches for words and phrases that it has specifically been trained to identify. Sometimes the true nature of these complaints is lost in translation, with unintended consequences.
Companies can either find themselves spending time and resources investigating complaints that should rather be dealt with as internal grievances or, perhaps even more worrying, risk missing serious complaints because of AI’s choice of words or phrases. Again, there is no substitute for human judgment.
AI advice and legal privilegeÂ
It is becoming more and more common for individuals and organisations to use AI to check the professional advice they receive from human advisers. However,
AI is not protected by privilege (as human advisers are), which can be problematic in disputes. Unlike legal advice obtained from a human lawyer, AI-generated advice might be discoverable and therefore have to be disclosed in future litigation.
South Africa has already had at least one case where input and output were found in court not to be protected by privilege. Our laws regarding privilege have changed over the years, becoming very complicated, and it is essential in a dispute situation to be extremely careful about what information is inputted into or obtained from AI.
Accountability if AI goes wrong
 As recent events have shown, AI hallucination is real and if it occurs, can be a major liability risk for businesses. The (now withdrawn) draft South African AI policy and framework talks about accountability for AI outcomes and specifically mentions creating accountability lines that lead to a named accountability official or entity.
This underscores the notion that using AI is not a way to avoid responsibility and accountability, and that human intervention is critical. It also speaks to the idea that accountability needs to be traceable.
But liability can be difficult to trace given the number of players involved in the AI value chain, including developers, deployers and users, any of which could contribute to something going wrong with AI and making it unclear where liability lies.
In light of this, the best course of action is to take steps to safeguard the business. Companies should set up AI governance structures, review and amend policies, train staff and monitor the use of AI’s impact on assets. From an external perspective, it is important to be intentional when contracting for AI and clearly allocate responsibilities and liabilities.
Last word
Business has moved from AI caution to adoption. The important thing now is to understand what AI tools do and how they are being used, and then to understand the risks and how to manage them.
This includes realising that adding an AI clause here and there to contracts and policies is not going to be enough. It also means understanding that the law is lagging and, until it catches up, aligning where possible with international best standards and the local guidance that is starting to emerge.
Finally, remember the human element.
