Identrics is a research and development company that creates award-winning AI solutions. Its solutions turn unstructured data into understandable knowledge and valuable insights.
The company obtains accuracy close to a human’s by utilising various approaches such as text mining, natural language generation (NLG), and natural language processing (NLP). Identrics offers insights about any volume of text in any major language using machine learning.
The CEO of Identrics, Vladimir Petkov, shared his perspective on the current state of artificial intelligence, including its ethical issues and what to expect next from Identrics.
Vladimir, you are well known for your thoughts and expertise regarding Ethics in AI. What is ethical AI, and what are some of the key ethical issues you think AI faces?
It is generally agreed upon that responsible AI means using AI in ways that do not harm people. Companies developed guidelines to ensure they are following specific rules that match the organisation's core values. These guidelines typically address issues such as liability or safety.
Several ethical issues need to be addressed as part of data science. These include prejudice based on race, political orientation, gender, hallucinations, factual inconsistencies, and hate speech. The data collection for machine learning models affects how well they work, which can lead to problems with bias.
Can you give us examples of AI and machine learning solutions in which you have integrated additional algorithms to guarantee they cover the Ethical AI principles?
The ethical principles that are most commonly known are:
-- Reliability & Safety
-- Privacy & Security
In Identrics, we are excited about creating AI and machine learning models, but at the same time, we understand the huge responsibility we have to develop an ethical technology. Therefore, our work is fine-tuned with additional algorithms to avoid biases, factual inconsistencies, hallucinations, hate speech, etc.
We can prevent the non-ethical behaviour of models by ensuring that data used to train them is unbiased, monitoring and controlling them, fact-checking and storing data appropriately, learning from past mistakes with model improvement processes such as the Human Information Processing (HITL) cycle, monitoring and detecting anomalies in production environments, and reassessing training data regularly.
In 2021, Identrics won a Gold Stevie® Award for Content Management Solutions with our Abstractive summarisation solution. Our technology is enriched with additional algorithms that ensure the high quality of the text and cover ethical standards.
In the Abstractive summarisation, we performed validation of all prompts, fact-preserving and fact-checking, and a “bad words” filter to sort toxic language. By the prompt engineering of the validation steps, absolutely all of the prompts were caught by the validation procedures. The feedback from users confirmed that the validated content was now correct and 100% inoffensive.
We believe in the future of technology, but we know that the role of humans is irreplaceable. AI and machine learning models are invented and created by people with the intent to make other people more flexible, creative, and effective. If we decide to give machines the ability to make autonomous decisions, we need to be very careful and have many safety systems in place to ensure that not a single person will get hurt.
Disinformation is also a topic nowadays. Do you think AI technologies can be used to detect it?
The world is facing a growing problem with disinformation and hate speech. Thanks to new technologies and social networks, fake news is more significant than ever before, aiding the spread of both disinformation and propaganda. In this case, technology can be both a stimulus and a solution to these problems.
Identrics has the experience of extracting information from open sources and turning it into knowledge in a short period. We monitor over 100 000 sources, collect data on more than 1.5-2 million articles daily, and use the information to deliver new insights to our clients.
We provide solutions to help you detect hate speech and disinformation in comments on blogs, news sites, forums, or other online communities. The latest projects we are working on with partners include machine learning models for detecting and alerting hate speech and propaganda.
The variations of this solution are:
-- hate speech for media and social platforms
-- style based disinformation detection
-- knowledge-based disinformation detection
-- propagation-based disinformation detection
-- source based disinformation detection
What to expect from Identrics in 2023?
We have been working on two major products for quite some time now. In 2023, we are planning to release both.
One of the projects is Kaspian, a big data delivery platform that deals with sizable data warehouses and organises them into data streams. What Identrics provides through Kaspian is a solution for a quick and simple setup procedure, full document access, and global monitoring. We are preparing to implement all of Kaspian’s features in the upcoming year.
Something else we have been developing is Topify. This is a product we have been working on for over a year. It is a tool for quantifying internet news that clusters information, provides a broad overview of new topical trends, and delves deeply into the parameters that drive them.