The nature of work has continuously evolved since the First Industrial Revolution. The exponential growth of data in this Fourth Industrial Revolution has led to the advancement of data-driven technologies such as Artificial Intelligence that can improve productivity and boost economic growth through the creation of new products and services.
AI has tremendous potential to improve society, however, since AI is still an emerging technology, we know that there are several AI systems which operate in unexpected or undesirable ways. Some prominent examples of not trustworthy AI that we have known created headlines globally – bias in online recruiting tools, in words associations, online ads, facial recognition technology, criminal justice algorithms to name a few.
As a tool AI can amplify both our best and worst decisions, and that we need to handle with care. Since most AI doesn’t deliver trusted predictions and insights consensus around which decisions are trustworthy are hard to come by.
Today machines are playing an essential role in our daily lives making or influencing some of the daily tasks as well as critical decisions affecting us and society at large. Machines are trained to harness the large volume of macro and microdata and their scale and statistical rigour promise unprecedented efficiencies.
Since, AI applications are programmed by humans and thus can be vulnerable and exposed to the biases of its programmers, or even may end up making biased judgments based on incorrect data. Apart from that, if the AI system priorities are not aligned with fairness, transparency, justice goals, even then the AI system can deliver negative outcomes.
Does this raise important questions about how to safeguard against bias and discrimination, as well as why is trust so important in the context of AI?
In the AI context, if you can’t establish trust, its adoption and usage will not yield results for the very reason AI too was created for. AI holds great promises as well as dangers, and hence everyone in the tech world today -organisations, consumers and regulators are especially concerned about how to build trust in AI to foster adoption or usage. This requires conscious effort to build transparency into AI models and the data that is being fed into the algorithms, giving users peace of mind that their data is being used appropriately to inform and improve decisions.
As humans, we use our cognitive and intuitive capacity whether to trust someone or not. We look at the facial expression, body posture, or the contextual background or may even compare with our memories or past experiences in a split of seconds.
Similarly, AI models are also prone to bias because they learn from the historical data that is fed to them, hence any bias in the data is either reflected or even amplified in the future predictions that the model makes.
It is thus extremely important to explore intended and unintended results of algorithms to identify, flag and mitigate associated impacts promptly. Some of the algorithmic bias exists due to their very nature like:
- Black box algorithms -which don’t provide any explanation
- Models are lacking diverse training data to handle special scenarios or critical situations.
- Models developed without having a diverse set of the team at the time of development and evaluation usually results in biased outcomes.
What steps should companies take to minimize algorithmic bias and develop Trustworthy AI for its customers?
AI ethics and governance should be embedded into AI applications and processes and should not be seen in isolation to be included post-AI application has been developed.
Organisations should focus on ethical principles that are most important for their sector and analyse how will they affect their business growth.
To create scalable and trustworthy AI, companies should carry out the AS-IS situational analysis to find the skill gaps and equip existing employees with the right education and tools that are aligned with ethical values.
All AI should depend on rich and diverse data sets, as well as inputs that are tested and validated. It’s important to process multiple data sources to run simulations and insights for algorithms, processing and validation. In cases with higher risk, ensure to run smaller tests or simulations before using them for broader public usage.
Carry out end to end audit and granular explanations for all models and predictions in the AI systems ensuring users understands how insights are produced at a granular level.
Through Machine Learning Ops ensure continuous refinement and performance evaluation for models. Highlight daily flags that require models to be refined or rebuild based on new data feeds and sources.
Manage the trade-off between full transparency which at times may be either impossible or not required due to data privacy challenges. To deal with such scenarios the input data can be explained, the outcomes from the system can be monitored, and the impacts of the system can be audited.
Maintain up to date model documentation like Factsheets so that stakeholders and users of AI Tools are aware of the model creation and changes were done to the model.
In conclusion, for algorithms consumers who aim towards the reduction of risk and associated bad outcomes, the adoption of mitigation techniques can help the creation of a pathway towards building trustworthy AI.
The superior ability of AI to recognise patterns creates serious potential ethical issues when it is used to make predictions about human behaviour. To ensure that predictive systems are not indirectly biased, all variables used to develop and train the algorithms must be rigorously assessed and tested. In cases with higher risk, it may be important to run smaller tests or simulations before using them on the broader public. In addition, the model itself should be assessed and monitored to ensure that bias does not creep in.
Join Our Telegram Group. Be part of an engaging online community. Join Here.
Subscribe to our NewsletterGet the latest updates and relevant offers by sharing your email.
Deepti has over 13 years of experience working with organisations undertaking transformational journeys, spanning many elements including IT, digitisation, customer experience, organisational change, agile at scale and cost reduction within the banking and financial services sector. She strongly believes AI and other emerging technologies are increasingly integral to the world we live in and organisations need to deploy these technologies at scale to remain relevant. Outside of work, she is passionate about Diversity and Inclusion within the STEM sector. Have been volunteering in various capacities with local non-profit organisations that encourage women to study and make a career in the STEM sector.