From Alibaba DAMO Academy's Top 10 Tech Trends: Mapping the Development Path for AI Enterprises
-
Recently, Alibaba DAMO Academy released its 2020 Top 10 Tech Trends, with "The Evolution of AI from Perceptual Intelligence to Cognitive Intelligence" topping the list.
What Are Perceptual Intelligence and Cognitive Intelligence?
The industry generally agrees that the main development directions of AI include computational intelligence, perceptual intelligence, and cognitive intelligence:
- Computational Intelligence: Primarily refers to rapid calculation and memory storage capabilities.
- Perceptual Intelligence: Involves sensory abilities such as vision, hearing, and touch. Both humans and animals possess these abilities, enabling interaction with the natural world.
- For example, autonomous driving relies on perceptual devices like LiDAR and AI algorithms to process driving information, while facial recognition payments use facial data for identity verification. Machines excel in active perception by leveraging DNN and big data.
- Cognitive Intelligence: Refers to the ability to "understand and think," handling complex facts and situations. Its development occurs in three levels:
- Language understanding
- Analysis and reasoning
- Personality and emotion
The Necessity of Evolving from Perceptual to Cognitive Intelligence
DAMO Academy's report highlights that AI has reached or surpassed human levels in perceptual intelligence (e.g., hearing, speaking, seeing) but remains in its infancy in cognitive intelligence, which requires external knowledge, logical reasoning, or domain adaptation. Cognitive intelligence is the key to unlocking AI's potential and expanding its industrial scale.
For instance, in public security, facial recognition only identifies individuals, but solving cases requires analyzing vast clues and deducing outcomes—tasks beyond perceptual intelligence alone.
Cognitive intelligence enables AI systems to move beyond simple data fitting, uncovering underlying patterns and causality. It allows machines to understand multimodal data, learn human-like "general representations," and acquire multisensory capabilities, delivering transformative industrial value.
The Path from Perceptual to Cognitive Intelligence
DAMO Academy asserts that cognitive intelligence will integrate natural language processing (NLP), cross-domain knowledge graphs, causal reasoning, and continuous learning to establish mechanisms for stable knowledge acquisition and expression. This will enable machines to comprehend and apply knowledge, achieving the critical leap from perceptual to cognitive intelligence.
NLP and knowledge graphs are foundational to cognitive intelligence. Companies leading in these areas will gain a competitive edge in the cognitive AI era.
Machine learning relies on text, but textual understanding poses significant challenges due to contextual variability and underutilized value. NLP translates natural language into machine-interpretable symbols, forming the basis for knowledge comprehension.
Once NLP is mastered, cross-industry and online-offline data integration becomes possible, paving the way for AI knowledge bases. Mastery of NLP and knowledge graphs thus unlocks greater value for enterprises.
Domestic Landscape and Current Developments in Cognitive Intelligence
Currently, cognitive intelligence companies primarily emerge from business transformations, falling into two categories:
- Big data analytics firms (e.g., MiningLamp, Baifendian)
- Speech/acoustic technology firms (e.g., iFlytek, AISpeech, Unisound, Kuaishangtong), mostly focused on speech and voiceprint recognition.
Example 1: iFlytek
- Perceptual Intelligence: Aims for highly usable sensory capabilities (e.g., speech, handwriting, image recognition).
- Cognitive Intelligence: Seeks breakthroughs to partially replace human mental labor (e.g., smart customer service, automated grading, human-computer interaction).
Example 2: Kuaishangtong
- Combines voiceprint recognition (perceptual "Biometric Brain") with NLP (cognitive "AI Empowerment Brain") to train expert systems across domains via speech and language, empowering society.
While perceptual intelligence plateaus, cognitive intelligence accelerates industrial upgrades. Though universal cognitive AI remains elusive, niche applications are already commercialized.
In public security, cognitive AI extracts criminal behavior patterns, developing predictive models to enhance threat预警 and risk perception (e.g., Kuaishangtong's Voiceprint Integrated Platform, MiningLamp's Big Data Analysis System).