2021 was a year of quantum leap for AI companies as they made huge headway with a record of $66.8B in funding. During the digital acceleration of the pandemic, the demand for AI-driven solutions across sectors, whether medical or metal, social media or security, has risen not just to improve product delivery but also to prepare against the onslaught of another uncertainty. 

Moving on to the year 2022, AI funding plummeted by 12% from Q4’21 to Q1’23. However, when looking at the overall funding trend, the year’s decline is quite less when compared to the entire venture which stood at the fall of 19%.

Factors such as rising regulatory risks, debilitated crypto currency prices and depletion in the amounts of megarounds of over $100M have created a funding winter in this sector. Additionally, some analysts believe that this is also a sign of a rise in acquisitions of AI start-ups by bigger corporations. 

With ChatGPT, metaverses, deepfakes and synthetic data, amongst many others, on the rise, the time is ripe for the era of machine learning. Check out these seven AI trends that a CB Insights report suggests to observe closely.

#1: Synthetic data - A road towards better privacy

Synthetic datasets are fake images, videos or tabular data. Companies use this invention in place of real-world data in order to abide by the GDPR and other privacy regulations while still facilitating data sharing and collaboration. 

The popularity of synthetic data is rising far and wide, from the medical to finance sector. The applications differ according to a company’s needs. For instance one startup developed synthetic data for biotechnology firm Illumina for medical research. The startup, Gretel, trained an AI algorithm to produce artificial genomic data with the help of real genotype and phenotype data, allowing innovation where real genomic data would have been subject to deep privacy concerns. 

Another example is in the telecom sector where up to 85% of real consumer data cannot be used given the absence of customer consent. In order to make behaviour analysis and prediction easier, one AI firm has generated synthetic customer profiles that are similar to real customers’ statistical patterns while also complying to GDPR.

#2: $67B AI chip market - a new race

A demand for specialised hardware accommodative of AI workflows that are compute-intensive has increased. This is true for both cloud data centers and edge devices like cameras.

Nvidia's domination of the AI chip market is ending as big corporations like Google and Amazon have launched their own custom chips and processors in Q4’21, and  startups are not far behind. Cerebras System, a startup with “the largest chip ever built” raised $250M in Q4 '21 at a $4B valuation. The chip has 2.6T transistors and 850,000 AI cores.

Interestingly, with the help of in-memory computing, various startups are experimenting with integration of AI processors and memory. The chips not just ease the process but also go on to affect the energy consumption. For instance, Samsung claimed that it was able to cut its energy consumption by 50% as they doubled the speed of a speech recognition neural network.

In this race, a revolution seems to be coming through various companies as they dump the traditional pathways of producing chips. They are transferring data through light instead of electrical signals, a turn towards photonic processes which are much faster than electrical.

#3: Natural language processing - shielding the online world

Meta has expended $13B on content moderation between 2016 and 2021. TikTok declared in 2020 that around 10,000 people were employed as content moderators. Importantly, both the companies have been sued by the employees working in this sector because of the psychological, traumatic harm it has done to them.

Companies are employing AI to identify any potentially toxic, harmful or discriminatory behaviour online. With gaming platforms becoming an infamous ground of hate speech and cyberbullying, corporations are turning to AI to combat the challenges of toxicity. According to an Anti-Defamation League study, 80% of the players in some of the most well-known games have gone through one or more kinds of harassment.

Startups like Spectrum Labs are creating natural language processing (NLP) platforms. They assert that the audio and the text-based content moderation efforts are reduced by 50% and detection of toxic conduct and actions is boosted by 10x.

Other startups like GGWP and Hive have raised fundings of $12M in 2022 and $50 M at $2B valuation in 2021 respectively.

#4: Deepfakes - figuring out the fallacies

From deepfakes of Vladimir Putin and Volodymyr Zelenskyy with respect to wartime misinformation in Ukraine and Russia, to ones of South Korean presidential candidate Yook Sun-yeol, it is getting harder to distinguish the real from the fake.

Paul England, Engineer at Microsoft Research Lab, said, “What we observe is that fake media is getting better and better, and the ability for computers to tell what's real and what’s fake, or users to tell what’s real and what’s fake, is rapidly tending to zero.”

To deal with the disinformation and scams that proliferate around deepfakes, huge corporations like Microsoft alongside Adobe and the BBC have launched their own initiatives that use tamper-proof metadata to authenticate media. Startups are also venturing into the field.

AI Foundation, a startup that launched the deepfake detection platform named Reality Defender with Microsoft, raised a total of $17M in 2020. It has also entered into partnership with the ABC, Homeland Security and Department of Defence.

Other methods such as use of cryptography and blockchain-based tech for photo and video identification, and reverse engineering have been developed by Truepic and Meta respectively. The former raised $27M in the year 2022.

These solutions do not, by any means, end the problem of deepfakes even though they do put a halt on the circulation of older versions. The ambiguity between real and fake, information and disinformation will continue to rise as deepfakes grow to become more elaborate and complicated. This will, in turn, push tech companies to find solutions to eliminate them.

#5: Augmented coding - replacing human programers

AI coding is bringing out faster results with the ability of algorithms to translate easy natural language commands into code. A job that was till now solely in the hands of humans is now succumbing to the effortlessness that AI provides. With this, marks the new dawning of the software development industry.

Microsoft and Google are running the show with the former’s acquisition of GitHub in 2021 and the latter’s collaboration with DeepMind in 2022. Startups are yet to rise in this field given the sector is still in its infant stage.

An important role that augmented coding is likely to play is making coding easy and accessible for non-technical users. A trend in the direction of no-code or low-code solutions should be awaited!

#6: Multimodal AI - a new search and content generation

The time is here for AI to make predictions about the video with the help of video data, about the text with text data. This new model of AI that is growing to comprehend multimodalities including videos, texts and 2D images are going to make the online world even holistic and simpler with its results.

With the launch of Imagen in May 2022, Google has made it possible to convert a text into an image while achieving a near to perfect alignment between the input and the AI generated image. It is further working for the future users to have the search for queries using a picture as an addition to their question.

Although initially employed majorly by academia, the multimodal AI is experiencing a warm welcome from more practical engines. For instance, Twelve Labs, a startup, raised $5M in March 2022 in order to index videos making them searchable as they extended the AI to learn from visual and audio data.

#7: AI first - end to end machine learning

Although a hard process to navigate, companies are ready to set the foundation for making their projects completely AI based, from raw data, production to management practices. As complex as it is, the market of “AI first” is increasing and diversifying rapidly with different vendors ready with solutions to different problems.

One of the most important challenges is the talent gap that occurs in different industries and departments with machine learning. Startups and big companies like Google have been working to bridge this gap with end-to-end AI development platforms and no-code data analysis tools.

This sector has been growing tremendously. For example, DataRobot, which has a valuation of $6.3B, made three acquisitions in one year so as to pull in a larger share of the enterprise AI market. With more companies signing up for machine learning, unicorn startups are bound to emerge to ease the process and battle the challenges.