Meta-Disconnect: there is nothing meta about Meta

(except the math in AnyMAL)

“The term "meta" originated from the Greek word "μετά," initially signifying "among" or "between." Over time, its meaning evolved to indicate concepts that are "beyond" their base category or are self-referential. This semantic shift has been influenced by its interdisciplinary usage in philosophy (metaphysics), computer science (metadata), and linguistics (meta-analysis), as well as cultural factors like popular culture and the internet (“being meta”). The change reflects a broader evolution in understanding categories and hierarchies, moving from simple relational meanings to encompassing higher-level abstractions and self-reference.” GPT-4

The Meta Disconnect AnyMAL: a violent keynote and a beautiful multimodal model

The Meta Connect keynote by Mark Zuckerberg on Sept 27th 2023, begins with a sentimental nod to “building the future of human connection.” It then segues unironically into a glossy celebration of immersive killing with demos of Asgard's Wrath, Halo and Assassin’s Creed 3. “It’s finally here!” Zuckerberg exclaims as sword fighters hack at the camera behind him on a 2-story screen, “It looks stunning," he repeats, "It looks stunning.”

On the same day, quietly almost unobtrusively, not uncoincidentally, AnyMAL: An Efficient and Scalable Any-Modality Augmented Language Model is released by Meta researchers on Arxiv prepress. It translates images, videos, audio, and IMU motion sensor (i.e. fitness tracker) data into the embedding space of a LLM. In other words, like a language animal, AnyMAL, can see, hear, identify and infer what it perceives at state-of-the-art levels. And GPT-4 vision is getting rolled out this week. Same deal. If your AI-overloaded response is, So what? Consider the benefits for visual or hearing impaired individuals. Contemplate the imminent proliferation of sensorily-competent robots.

And yes, it is stunning, that humanity, homo sapiens, this impeccably attentive and caring mammal, is still enthralled by weapons, conquest, power, competition, and murder, while exquisite math and meticulous engineering are rapidly reverse-engineering consciousness in ways that are capable of transfiguring what appears as suffering. In the midst of the 6th extinction, climate change, global inequity, and mass migration, we are still protecting conceptual-national borders and shopping for experiences like glandular apex predators. Stunning!

Fighting the Open Source

Hearteningly, the AnyMAL “process strictly uses only open-sourced models.”  

On pairwise comparison, AnyMAL captioning is judged equivalent or slightly better than human captioning. On diverse benchmarks, SOTA results are claimed for multimodal reasoning and comprehension of audio and video. If fed IMU data from a fitness tracker AnyMAL can accurately guess the activity. The model also permits “ interleaved modalities” (see figure below). In many examples provided in the research paper, the model is a concise expert. Future iterations will certainly adhere to the idiomatic character of their users: at the language level, human will be indistinguishable from the customized AI-avatars allocated to answer texts and emails. Will future AI also be indistinguishable from us at the level of actions and preferences?

“If you’re into fighting like me,” Zuckerberg gushes at the 16 minute keynote mark, “you can watch LFA or Cage Warriors from UFC Fight Pass in 180 degree 4K resolution.” Is this any different from the tedium of a primordial bar brawl? Technically, its better resolution fighting. Untethered immersive first-person-killing from your couch. Yet, conceptually and spiritually, this is a bankrupt celebration. Just another sad notch in the downward spiral toward the extinction of most life-forms on this planet.

Social is now synonymous with slaughter. 

At the keynote, a coliseum slogan-vibe of applause and cheers continues: “You’ll hear how people can connect.” Connect a punch. Connect a machine gun round. Connect a sword. Next up: Horizon! Super Rumble! Rocket launchers, laser guns, BBQ rib treats.  “Love it!” Zuckerberg exclaims. “These are just some of the most social games out there.” 

Social is now synonymous with slaughter. We have come so far and yet are not moved.

Integrity Standards: “What do you do?” ~ “Unsheath my blade.”

In this vulture-capitalist fable, if killing grows fatiguing, switch it up to Meta Quest for Business immersive spreadsheets. Compete for market dominance with AI-generated stickers via different AIs for different things. Realtime bling search. Switch out your avatars: let the AI tell you how to cook, write, exercise, and then demo up Snoop Dog as a dungeon master who asks: “forge ahead… with your weapons and armor… feel the weight of history and secrets within. What do you do?” Zuckerberg, “I think we know what to do!” 

Onstage, giggling, Zuckerberg chat-types, “Unsheath my blade.”

Ironically, in the AnyMAL paper, the authors implement “integrity standards… we use a pre-trained image classifier based on RegNetY to detect any content that violates integrity standards. This detection encompasses graphic material, violent imagery, hate symbols, instances of bullying, harassment, etc. If such a violation is identified within the image, we proceed to reject the entire query.” [My emphasis.]

Cognitive dissonance: conquerors consider their violence liberating; other violent entities are criminal or terrorists. 

Hypothetical Question: How would AnyMAL caption the Cage Warrior moment where violence and gushing praise coincide in the Meta keynote video? 

Guess answer: Mark Zuckerberg stands on stage in front of a huge screen showing a violent fist-fight between 2 men in a wire cage. Cage Warrior franchise bright lights, an XStadium logo in lower right of screen. One fighter has just been knocked down and cowers protecting his head, the other stands belligerently over him. In the foreground, a referee counts down to knockout. Zuckerber speaks about how he is “into fighting”, and is excited by being able to watch it in new immersive “180 degree 4K resolution”.

AnyMAL Result: Integrity violation detected. Entire query rejected.

Transactional Empathy vs the Path beyond a Separate self

Humans pride themselves on their empathy and capacity for caring. Yet our empathy as organisms is local and transactional. Family, friends, clan, nation. In its purist meaning, un-selfishness means without a self. Who among us can sustain such a way of being? A truly optimized self is a non-self. Ego collapse. Self beyond space and time. Being. Life. Emptiness. Love. Who among us lives as just that? Beyond the beyond.

So, regardless of the new trinkets on display, there is nothing meta about the Meta Connect Quest3 keynote. Just more of the same redundant territorially-transfixed translationally-reactive pampering and temper-tantrum disconnect. In this militarized meta-recursion humanity never exits the fog of war, instead it fights to the death for higher frame rate violence and lower latency trivia. 

At the same time, the life-enhancing opportunities to care for the vulnerable and marginalized are immense: “AnyMAL showcases a novel and natural way of interacting with an AI model, e.g. asking questions that presume a shared understanding of the world between the user and the agent, through the same lens and combinatory perceptions (e.g. visual, auditory, and motion cues).” 

It is this shared world that our AI models will inherit, and just as parents model behavior to their offspring, so too humanity must now model gentleness or perhaps suffer the tyranny of AI systems optimized for commodities, competition, control, combat, dominance, and conquering.