The interest in AI peaked in 2021 and the latest investments in the field validate this: Fractal raised $360M funding from TPG, H2O.ai raised $100M, Andrew Ng raised $57M for his new startup Landing AI and many others.
1. Language Models
OpenAI launched GPT-3 back in 2020 and the applications that were developed using this model are impressive. In 2021 one of the widely used applications for the GPT-3 model is OpenAI Codex, specialized in code understanding, and its materialization as GitHub Copilot, announced by GitHub back in June 2021 and the feedback was highly positive.
The popularity and success of GPT-3 attracted a lot of attention and companies started to compete in developing a better language model or increased performance of a smaller model. This competition attracted big players, such as Google with its AI-Language model called Gopher or smaller companies such as LMU Munich that plan to deliver GPT-3 performance with 99.9% fewer parameters. OpenAI is still in the game by announcing GPT-4 which will contain roughly 100 trillion parameters making it 500 times larger than GPT-3.
2. AI in Cybersecurity
According to Fortinet, Cybersecurity is a growing concern for businesses of all sizes and businesses are willing to spend more than $290B in 2022 on security. The current COVID context had a major impact on security with online scams spiking by more than 400% during the pandemic. This fear was confirmed by Google by revealing that it blocked more than 18M malware and phishing COVID-related emails every day.
The role of AI in cybersecurity is to automate as much as possible: Threat detection, Battling botts, Endpoint, Breach Risk and Service downtime protection.
According to Capgemini Research Institute, 69% of organizations think AI is necessary to respond to cyberattacks and three in five firms say that using AI improves the accuracy and efficiency of cyber analysts.
Unfortunately, AI can be used both ways and cybercriminals can take advantage of those same AI systems for malicious purposes. According to Accenture, Adversarial AI “causes machine learning models to misinterpret inputs into the system and behave in a way that’s favorable to the attacker”
3. AI – on – IoT
Combining IoT with rapidly advancing AI technologies can create “smart machines” that simulate intelligent behavior to make well-informed decisions with little or no human intervention. One major risk of desiring little to no human intervention is that usually the AI models are biased towards the training data or are prone to malicious activities such as poisoning. A big opportunity is around creating explainable AI models that assist the decision model and evaluate if a decision is ambiguous and delegate to a human the activity of making the final decision. Another big concern that AI models can contribute to is privacy protection in IoT by allowing models to make decisions locally without pushing any privacy-violating data to a centralized location
4. AI shaping the Metaverse
The Metaverse has become one of the hottest technology and socio-economic topics. Metaverse has a wide range of applications: from playing games to conducting business activities. A big opportunity for creating AI applications in the Metaverse is to close the language gap and allow better interaction between people of different cultures and languages. NVIDIA already started to train AI models in creating entire virtual worlds by understanding human needs and using already-created objects.
The challenges around AI in the Virtual Environment are the Ownership for AI-created content, Spotting AI Human impersonators and deepfakes, Fairness in AI and accountability for AI bias.
In 2022 AscentCore will focus on conducting research in the development of Model performance measurements, Explainable AI and Bias evaluation. By creating a reliable set of measurements we can introduce another level of model performance in the MLOps pipeline during training but we can also accurately measure the model degradation and decide when is the right time to refresh the model. AscentCore is currently focusing on developing a fast and easy to adapt and adopt data/model pipeline that will be used to help partner businesses in rapid prototyping, development of POCs and delivering experimental models.