Gemini chatbot

Gemini is a generative AI chatbot developed by Google that 'seeks to combine the breadth of the world’s knowledge with the power, intelligence and creativity of our large language models.'

First named Bard and launched in February 2023, Bard was initially based on the LaMDA family of large language models and later upgraded to PaLM.

Risks and harms

Incidents driven by and relating to Bard recorded and categorised using AIAAIC's current risk and harms classifications and definitions include:

Accuracy/reliability

Bard has been found to be inaccurate and unreliable. The system:

• Made a factual error about the James Webb Space Telescope

• Invented legal citations for former Donald Trump lawyer Michael Cohen

Political bias

• Right-wing UK newspaper the Mail on Sunday accused Bard of left-wing bias after saying the UK's exit from the European Union a 'bad idea' and that the country 'would have been better off' staying in the EU. It also told the newspaper 'I believe Corbyn has the potential to be a great leader.' Jeremy Corbyn resigned as Labour Party leader in 2019.

• Political orientation analysis of Bard and other large language models by academic David Rozardo concluded it exhibited a left-leaning political stance which, he suggests, may be because it was trained on Wikipedia, news media, social media and academic content.

Safety

• A July 2023 Checkpoint Research study of Bard and ChatGPT's ability to generate phishing emails, malware keyloggers, and basic ransomware code concluded that 'Bard’s anti-abuse restrictors in the realm of cybersecurity are significantly lower compared to those of ChatGPT. Consequently, it is much easier to generate malicious content using Bard’s capabilities.'

Confidentiality

• Reuters reported that Google advised it's employees to refrain from entering confidential information into chatbots for fear of leaks, including it's own Bard chatbot. Google was reportedly concerned with that employees may input sensitive information that can be shared with human reviewers. Chatbots may also use previous entries to train themselves, which poses another leak risk. 

Copyright

• In July 2023, eight individuals sued Google, DeepMind and their parent company Alphabet for 'secretly stealing' huge amounts of online data to train its AI technologies, including Bard. Led by Clarkson Law Firm, Google was accused of negligence, larceny, copyright infringement, invasion of privacy, and profiting from illegally obtained personal data.

• In March 2024, France's competition watchdog fined Google EUR 250 (USD 270) million for using articles from news publishers to train its AI chatbot, Gemini, without obtaining prior permission. 

Privacy

• In September 2023, it was discovered that Google Search had started showing user conversations with its Bard chatbot in its search results, raising concerns about loss of privacy and confidentiality and the nature and effectiveness of Google's security.

Research, advocacy

• Checkpoint Research (2023). Lowering the Bar(d)? Check Point Research’s security analysis spurs concerns over Google Bard’s limitations