Unintended consequences of AI's rapid advancement

AI is often called a double-edged sword as it can be used for causing as much harm as good. The rapid adoption of AI in recent years has exposed numerous issues, both inherent to AI models and external to them. For instance, the lack of homogeneity in data to train AI models has led to concerns about the exclusion of marginalized and non-English-speaking populations from the benefits of AI. Also, the use of human-generated material to train AI models without proper authorization from their creators has put the livelihood of journalists, artists and writers at risk. 

Not to forget, AI has led to a spurt in deepfake online scams and is expected to make millions of jobs obsolete. 

Here are some of the issues brought on by the hurried and unchecked development and adoption of AI: 

Rise in AI-generated online scams

The AI hype is also getting a lot of unwanted attention from cybercriminals who are increasingly using AI generated imagery for online scams. Generative AI-powered deepfake technology can create highly realistic videos and audio recordings of real people, making it easier for scammers to impersonate them and trick their family, friends and fans. AI can also craft highly personalized and convincing phishing emails to steal data and money. According to a survey by cybersecurity firm McAfee, 45% of online shoppers in India have been duped by deepfake shopping scams or know someone who has. Out of those affected, 56% lost money, with 46% losing over Rs 41,500 to such scams.

Risk of discrimination by AI

If an AI is trained on data that has inherent bias towards certain communities or if a certain group is not properly represented, the AI models can make errors. This can lead to discrimination and harassment of marginalized communities, especially when used for law enforcement and identity verification. In February 2022, a pregnant black woman was arrested by Detroit Police for robbery and carjacking based on a false face recognition match. The woman is now suing the city and police officials for the wrongful arrest.

Scarcity of Indic language data for training AI

There are 60 Indian languages spoken by over 10 lakh people and 125 languages with more than 1 lakh speakers. Yet, many of these languages don't have any digital data. Even widely spoken Indian languages such as Hindi, Bengali, Telugu, and Tamil, have a smaller digital footprint compared to English. Insufficient Indic language data for training can affect the AI model’s performance and accuracy. To address this gap, the Indian government has built a digital database called Bhashini to collect and make local language data available to enterprises and researchers.

Impact on jobs

While AI will boost productivity,  it is also expected to disrupt the job landscape and make many jobs obsolete. Consulting firm McKinsey estimates AI will replace 12 million jobs by 2030. This will lead to disruption, but experts believe that AI is also expected to create a lot more jobs that it will take away. The World Economic Forum is expecting AI to create 90 million jobs globally. However, it is believed that most of the jobs created by it will require workers to learn AI skills to develop or use AI models.

Unauthorized data harvesting

In recent months, several big firms including Google, Meta and OpenAI have been taken to court for using human-generated material, including copyrighted work of artists, journalists, and authors without their consent. Last December, NYT sued OpenAI and Microsoft for using their news articles without permission to train ChatGPT.  In May, actor Scarlett Johansson threatened legal action against OpenAI for copying and using her voice in ChatGPT’s new voice assistant Sky, despite her refusal to license it. 

All this backlash has led AI firms such as OpenAI and Perplexity to partner with media publications to use their copyrighted material for generating news summaries with proper attribution to the original article.  

AI’s hidden carbon cost

The rising demand for AI chatbots has led to a spike in power consumption of data centres that house the GPUs used to train and run AI models. A study by Semi Analysis shows that  AI will drive data centres to consume 4.5% of global energy consumption by 2030.  Google’s carbon emissions have already increased by 50% in five years due to AI energy demand. Many of the firms are now exploring nuclear and other forms of renewable energy to meet their climate goals.

Image credit: Pixabay