Fireside Chat: Blockchain x AI

DFINITY
2 Aug 202426:23

Summary

TLDRトレイシー・トラクサーとドミニック・ウィリアムズによるインタビューの要約です。トレイシーはDesity Foundationの機関関係責任者、ドミニックは創設者兼最高科学者であり、AIとブロックチェーンの組み合わせの将来性について語ります。ドミニックは、ブロックチェーン上で動作するAI、特にスマートコントラクトに強い信念を持ち、AIトークンやディープラーニングGPUネットワークには慎重な姿勢を示します。彼らは、AIがスマートコントラクトとして実行されることで得られる耐改ざん性、不停止性、自律性という利点を強調し、実用的な応用例を紹介します。また、インターネットコンピューター上でAIを実行する技術の進歩と、今後の大きな技術革新について触れています。

Takeaways

  • 😀 トレイシー・トラクサーはDesity Foundationの機関関係責任者であり、ドミニック・ウィリアムズはその創設者兼チーフサイエンティストです。
  • 🔒 彼らはAIとブロックチェーンの組み合わせに非常に前向きで、特にスマートコントラクトとして実行されるAIに注目しています。
  • 💡 AIをブロックチェーン上で実行することで、改ざん耐性、不停止性、オプションで自律性を獲得するという3つの重要なプロパティが強調されています。
  • 🚫 トークンやディープGPUネットワークなどのAIに対するトークン化は、彼らにとってはあまり魅力的ではありません。
  • 🤖 AIがスマートコントラクトとして実行されると、セキュリティ面での利点が示されており、ハッカーの標的になる可能性があることから保護されます。
  • 🏦 実用的な例として、AIスマートコントラクトが法律事務所の効率を高める方法が説明されています。
  • 💼 スマートウォレットの例では、AIが組み込まれたウォレットがユーザーの資産をより安全に保護し、自律性をもたらす方法が示されています。
  • 🔮 インターネットコンピュータは、AIをスマートコントラクトとして実行する唯一のブロックチェーンであり、これは他のプラットフォームとは異なる価値を提供しています。
  • 🎭 最近のデモでは、顔認識を実行するAIスマートコントラクトのパフォーマンスが向上し、その進化が示されています。
  • 📈 大規模言語モデルをブロックチェーン上で実行することは、インターネットコンピュータの次の大きな目標であり、これにはWASMの64ビットへの移行が必要とされています。
  • 🌟 インターネットコンピュータには、AIスマートコントラクト以外にも別の大きな進化が計画されており、これは技術の未来を変える可能性があると言っています。

Q & A

  • トレイシー・トラクサーはデジティ基金会のどのような役職を務めていますか?

    -トレイシー・トラクサーはデジティ基金会の機関関係部門の責任者を務めています。

  • ドミニック・ウィリアムズはデジティ基金会でどのような立場にありますか?

    -ドミニック・ウィリアムズはデジティ基金会の創設者であり、最高科学者です。

  • AIとブロックチェーンの組み合わせについて、ドミニックはどのような見解を持っていますか?

    -ドミニックは、ブロックチェーン上で動作するAI、特にスマートコントラクト形式のAIに非常に楽観的です。しかし、AIのためのトークンやディープラーニングGPUネットワークについては賛同していません。

  • トークン経済がAIにどのように影響を与えるとドミニックは考えていますか?

    -ドミニックは、多くのAI関連のトークンが実際には「スネークオイル」のような叙事を売りに来ており、ブロックチェーンとAIの進歩を促進していないと見ています。

  • AIがブロックチェーン上で動作することの利点は何ですか?

    -AIがブロックチェーン上で動作することで、改ざん不可能性、止められず、オプションで自律性という利点があります。

  • スマートコントラクトとして動作するAIの「改ざん不可能性」とは何を指しますか?

    -「改ざん不可能性」とは、ソフトウェアが常に正しいロジックと正しいデータに対して実行されることが保証されていることを指します。つまり、サイバー攻撃に対して免疫力があることを意味します。

  • 「止められない」AIスマートコントラクトとは何を意味しますか?

    -「止められない」とは、AIスマートコントラクトが常に実行されることが保証されていることを指しており、外部からの妨害や停止命令に影響されないことを意味します。

  • AIスマートコントラクトの「自律性」とはどのような性質ですか?

    -「自律性」とは、スマートコントラクトが誰によっても変更または更新されず、またはDAOによってのみ変更可能であることを指します。これは、例えばスマートウォレットのようなアプリケーションにおいて、コードが悪意のある変更から守られるため非常に重要です。

  • ドミニックは最近のデモでどのようなAIの機能を示しましたか?

    -ドミニックは最近のデモで、顔認識機能を持つAIスマートコントラクトを紹介しました。これは、AIスマートコントラクトのパフォーマンスが向上したことを示すだけでなく、多くのアプリケーションにおいて顔認識技術が有用であることを強調しています。

  • ブロックチェーン上で動作するAIが未来に何をもたらすとドミニックは考えていますか?

    -ドミニックは、将来的には数百万人から数十億人のカスタムAIモデルが存在し、その多くがスマートコントラクトとして動作するようになると考えています。これは、人々がAIをより信頼し、その利点を活用するためです。

  • インターネットコンピュータの新しい機能について、ドミニックはどのような予告をしていますか?

    -ドミニックは、インターネットコンピュータに9月初めにベータ版で公開される新しい機能があると予告しています。この機能は技術の未来を変えるほどのゲームチェンジャーであり、インターネットコンピュータが提供できるユニークな機能であると述べています。

Outlines

00:00

🤖 AIとブロックチェーンの組み合わせの強み

パラグラフ1では、Desity Foundationの創設者で最高科学責任者であるDominic Williamsが、AIとブロックチェーンの組み合わせに対するその期待とその理由について語っています。特に、ブロックチェーン上で動作するAI、すなわちスマートコントラクト形式のAIに注目しています。トークンやディープGPUネットワークに対しては懐疑的な姿勢を示しており、ブロックチェーン技術がAIにどのように基本的なレベルで価値を加えるかについて語っています。

05:01

🚫 AIトークンとディープGPUネットワークの限界

パラグラフ2では、AIトークンの多様性とその多くが実際には単なる偽物販売に等しいというDominicの見解に触れています。また、ディープGPUネットワークの課題についても議論しており、大規模データセンターでのGPUの利便性と価格面での利点を強調しています。Elon MuskによるNvidia GPUの購入など、中央集約型AIクラウドの強みを説明しており、ブロックチェーン上でAIを実行することで得られる独自の利点を探求しています。

10:02

🛡️ AIスマートコントラクトの三大特性

パラグラフ3では、AIをスマートコントラクトとして実行することで得られる三大特性:改ざん不可能性、止められない性、自律性について解説しています。これらの特性がAIのセキュリティや信頼性、そして特定のアプリケーションにおいての重要性について説明されており、具体的な例として法律事務所のAIアプリケーションが紹介されています。

15:02

👨‍💼 法律事務所におけるAIスマートコントラクトの活用

パラグラフ4では、法律事務所がAIスマートコントラクトを導入することで、裁判の準備期間を劇的に短縮し、重要情報を見落とさずに効率的に業務を遂行できるようになった例が紹介されています。AIの改ざん不可能性と止められない性が、法律事務所にとっての重要性について強調されています。

20:03

💼 Web 3におけるスマートウォレットと自律性

パラグラフ5では、Web 3の文脈でAIスマートコントラクトがどのように活用されるかが議論されています。スマートウォレットに組み込まれたAIが、ユーザーの指示に応じて自動的にトランザクションを行う様子を想像し、自律性の重要性が強調されています。また、イーサリアムのSolidityスマートコントラクトを監査するAIスマートコントラクトの例も紹介されています。

25:03

🔍 顔認識デモとAIスマートコントラクトの進化

パラグラフ6では、顔認識デモを通じてAIスマートコントラクトのパフォーマンスがどのように向上したかが示されています。Desity Foundationの取り組みや、WebAssemblyへの貢献など、技術的な背景が触れられています。さらに、インターネットコンピュータ上でより大きな言語モデルを実行する計画も語られており、その実現がAIとブロックチェーンの進化において重要な意義を持つとしています。

🎉 インターネットコンピュータの次の大きな進化

パラグラフ7では、インターネットコンピュータの今後の展開に期待が高まっています。Dominic Williamsは9月上旬にベータ版で公開される新しい機能について触れ、それが技術の未来に与える影響が多大であると強調しています。この新しい機能はインターネットコンピュータ独自のもので、AIとブロックチェーンの密接な関係をさらに示すものであるとしています。

Mindmap

Keywords

💡ブロックチェーン

ブロックチェーンとは、データのブロックがリンクされた分散型台帳のことです。このビデオでは、ブロックチェーンがAIと組み合わせることで、新たな技術の進化をもたらすと強調されています。例えば、AIをブロックチェーン上で動作させることで、セキュリティや信頼性の高いアプリケーションが実現可能になるという点が説明されています。

💡AI

AIとは、人工知能の略で、機械が人間のように学習し判断する能力を持つ技術です。ビデオでは、AIをブロックチェーン上で動作させた「スマートコントラクト」として、新たな機能性やセキュリティを提供することができると語られています。

💡スマートコントラクト

スマートコントラクトとは、ブロックチェーン上で実行される自動化された契約であり、特定の条件が満たされた時に契約の実行を保証します。ビデオでは、AIをスマートコントラクトとして実行することで、その不変性や実行の確実性を高めることができると説明されています。

💡カスタムAIモデル

カスタムAIモデルとは、特定の目的やニーズに合わせて設計された人工知能モデルのことです。ビデオでは、将来的には数百ものカスタムAIモデルが存在し、その多くがスマートコントラクトとして実行される可能性があると予測されています。

💡分散型AI

分散型AIとは、複数の場所に分散して動作するAIのことを指します。ビデオでは、分散型ネットワーク上のGPUで動作するAIモデルについて議論されており、その利点と課題について触れられています。

💡データセンター

データセンターとは、コンピュータシステムを集中的に運用する施設で、大きなストレージ容量や処理能力を提供します。ビデオでは、データセンターがAIモデルを実行する際の利便性やコスト効率について言及されています。

💡トークン

トークンとは、デジタル通貨や資産を表す単位であり、ブロックチェーン上で取引されることがあります。ビデオでは、AIトークンについて批判的な視点から語られており、その多くが実際の価値を提供していないとされています。

💡GPU

GPUとは、グラフィック処理ユニットの略で、画像やビデオの処理に特化したコンピュータチップです。ビデオでは、AIモデルのトレーニングや実行にGPUが使用され、その分散型ネットワークでの課題についても触れられています。

💡セキュリティ

セキュリティとは、システムやデータに対する不正アクセスや攻撃から保護する能力を指します。ビデオでは、AIがブロックチェーン上で動作することで、セキュリティが強化され、ハッカーからの攻撃から守られることが示唆されています。

💡自律性

自律性とは、システムや機械が人間の介入なしに独立して動作する能力です。ビデオでは、スマートコントラクトが自律的であることにより、信頼性や安全性が高まり、そのコードが改ざんされることがないことが重要であると強調されています。

💡インターネットコンピュータ

インターネットコンピュータとは、分散型のプラットフォームで、サーバーレスコードを実行し、スマートコントラクトを可能にします。ビデオでは、インターネットコンピュータ上でAIを実行することで、新たな技術の進化やアプリケーションが実現されると示唆されています。

💡ファシアルリコグニション

ファシアルリコグニションとは、顔認識技術を指し、ビデオではAIスマートコントラクトとしてデモンストレーションされた機能の1つです。そのデモンストレーションは、AIスマートコントラクトのパフォーマンス向上を示す重要な例として挙げられています。

💡WebAssembly

WebAssemblyは、ウェブブラウザ上で実行される高速なバイナリコードフォーマットです。ビデオでは、インターネットコンピュータがWebAssemblyを利用し、AIスマートコントラクトの実行環境を改善する計画が語られています。

💡ニューラルネットワーク

ニューラルネットワークとは、人間の神経系をモデル化したAIアルゴリズムのことです。ビデオでは、ニューラルネットワークがAIの進化と機能性に寄与し、将来的にはインターネットコンピュータ上でより大きなモデルが実行可能になると期待されています。

Highlights

Tracy Traxer作为Desity Foundation的机构关系负责人,与创始人兼首席科学家Dominic Williams就AI、区块链以及AI与区块链的结合进行了深入讨论。

Dominic Williams对AI在区块链上运行,特别是以智能合约形式的AI表现出极大热情。

Desity Foundation目前专注于AI与区块链结合的研发工作。

对AI代币和深度学习GPU网络持保留态度,认为这些并非推动区块链和AI发展的关键。

区块链技术在AI领域的核心价值在于其为AI提供了不可篡改、不可阻挡和可选的自治特性。

智能合约形式的AI具有不可篡改性,能够抵御网络攻击,保证执行正确逻辑。

智能合约形式的AI能够保证持续运行,不受数据中心故障等因素的影响。

自治智能合约意味着除非通过DAO投票,否则代码不能被修改或更新。

智能合约在法律行业中的应用示例,展示了AI如何大幅提高律师准备案件的效率。

如果智能合约AI被黑客攻击,可能会导致敏感数据泄露,对客户和律所造成严重后果。

智能合约AI的不可阻挡性对于依赖AI的业务至关重要,确保服务始终可用。

Web 3领域中,智能合约钱包的自治性对于防止开发者后门操作至关重要。

展示了智能合约AI在面部识别方面的应用,突显了其强大的性能和应用潜力。

智能合约AI的面部识别演示显示了其相较于早期演示在性能上的显著提升。

Desity Foundation在智能合约AI领域的持续研发,包括对WebAssembly的改进和优化。

大型语言模型在区块链上的运行,预示着与区块链对话的未来可能性。

互联网计算机执行环境的升级将允许更大型智能合约的运行,包括大型语言模型。

Dominic Williams预告了互联网计算机即将到来的另一项重大技术进步,虽然细节未透露,但强调其将改变技术的未来。

Transcripts

play00:00

hi everyone my name is Tracy traxer I'm

play00:02

the head of institutional relations at

play00:04

the desity foundation and I'm incredibly

play00:07

pleased to be joined by our very own

play00:08

founder and chief scientist Dominic

play00:10

Williams today as we head into a bit of

play00:13

a chat about AI the overinflated the

play00:16

under understated and all things

play00:18

blockchain and AI very much looking

play00:20

forward to this conversation well thank

play00:22

you for having me I think this is a

play00:24

super super super important topic and

play00:26

and I'm look forward to Shar what I can

play00:28

now of first Dom you been a big

play00:30

proponent of the AI and blockchain

play00:33

combination for quite a while now and

play00:34

it's driving obviously a lot of the R&D

play00:37

effort that the definity foundation is

play00:38

focusing on at the moment as well can

play00:40

you tell me a bit about what it is that

play00:43

makes you so bullish about this

play00:46

combination well we're specifically very

play00:48

bullish about AI that runs um on on on

play00:52

the blockchain itself in the form of

play00:54

smart contracts um we're bullish about

play00:56

something else too extremely bullish

play00:58

about something else too I I can't talk

play01:00

about today but we'll come back to that

play01:02

um our Viewpoint is a little bit tical

play01:06

so we're not so bullish about um tokens

play01:09

for AI and and we're not so bullish um

play01:12

about you know deep in GPU networks um

play01:16

or albe they got potential we're we're

play01:19

really um very focused on running AI on

play01:23

on the blockchain all right we'll head

play01:25

into a bit of a deeper discussion about

play01:27

AI Asar contracts in a second but you

play01:30

just mentioned something that has piqu

play01:31

my interest so uh you said that you

play01:34

don't feel like tokens for AI and deep

play01:37

in networks and gpus Etc will be huge

play01:40

why um well look start starting with

play01:44

with topens for AI um you know you can

play01:47

see there are hundreds of them uh you

play01:49

know on coin market cap and you know

play01:50

we've dug in because we want to

play01:52

understand the space um my my general

play01:55

impression personally is that that a lot

play01:58

of them are just you know pedling sort

play02:00

of snake oil narratives um in order to

play02:03

collect you know money sell tokens

play02:06

essentially to to dents who don't have

play02:08

the full picture um and you know I don't

play02:11

think that really advances the cause of

play02:13

of blockchain and AI um and you know

play02:17

there are a lot of blockchains right now

play02:18

who who've got big treasuries and and

play02:20

what they're doing is you know going

play02:22

around investing in these Web Two um AI

play02:25

projects and saying look in return for

play02:27

our investment you got to agree to

play02:29

credit token on on our blockchain and uh

play02:33

you that indirectly creates a lot of

play02:35

hype and then um they can say our

play02:37

blockchain is the blockchain for AI and

play02:39

they can sell more of that base token

play02:40

and so on and the flywheel keeps

play02:42

spinning kind of thing um you know

play02:44

that's not really what we're about um

play02:47

we're we're much more interested

play02:48

interested in you know how blockchain

play02:50

technology can add value uh to to AI at

play02:54

a much much more fun fundamental level

play02:58

um with with with respect to you know

play03:02

deepen um networks for gpus um I think

play03:05

you know these things can work but there

play03:07

are some issues we haven't seen um

play03:10

resolved so you know we think in the

play03:14

future that you know there are going to

play03:16

be millions or billions of custom AI

play03:20

models running right um and people like

play03:24

Mark Zuckerberg have you know made

play03:27

similar claims and you know they hope

play03:28

that uh you know for example llama will

play03:31

will be which is an open source AI model

play03:34

will be used to create a lot of those

play03:36

custom models um and the question is you

play03:39

know where are they going to run right

play03:41

and

play03:43

um you know you might think that well

play03:45

there's all these custom models why

play03:47

won't people run them on on on a

play03:50

decentralized you know network of

play03:53

gpus and to to understand some of the

play03:56

challenges uh you have consider that you

play03:59

know when people choose where to run

play04:01

these models they're going to look at

play04:03

you know capability stroke features and

play04:07

and price so when you're able to run

play04:11

gpus inside a single huge data center

play04:15

you're able to use a lot of uh special

play04:18

like Hardware um to to connect them um

play04:21

for example so that gpus can share their

play04:24

memory

play04:25

directly and you can't do that in a you

play04:29

know

play04:30

a a decentralized Marketplace so easily

play04:32

so you know if you're renting out the

play04:34

GPU on someone's gaming computer well

play04:37

you know that's just a sort of

play04:38

Standalone GPU and in fact like often

play04:41

times for things like training people um

play04:44

you know want to use gpus that are

play04:46

clustered together um so the feature set

play04:50

of you know centralized AI clouds

play04:54

created by people like Amazon and

play04:55

Microsoft or newcomers like V Ai and so

play04:58

are going to be much richer

play05:00

um and that that's so so in that case

play05:04

you know people will tend to favor the

play05:06

centralized a clouds over a deepend

play05:09

network and and it's also the case that

play05:11

they're going to be able to rent these

play05:14

gpus um for far less

play05:17

because they have these um economies of

play05:20

scale and and benefit from bulk um

play05:23

buying power so I don't know if uh

play05:26

anyone noticed uh recently um you know

play05:30

it came out that Elon Musk had

play05:31

redirected a $500 million shipment of

play05:35

Nvidia gpus from Tesla 2X and

play05:40

xai

play05:41

and I think that you know Elon musk's

play05:44

buying these Nvidia gpus in $500 million

play05:48

Clips right um and that gives you an

play05:50

idea of the scale of what's going on um

play05:54

and if you're buying uh you know your

play05:56

gpus from Nvidia directly from Nvidia in

play05:59

500 $ million Clips you've got amazing

play06:02

purchasing power right and amazing

play06:03

negotiating power so you're going to get

play06:05

a fantastic discount so you know uh

play06:09

these organizations running these

play06:10

centralized um specialist AI

play06:13

clouds um are paying less for the

play06:15

underlying Hardware um they're able to

play06:17

consolidate all of these gpus in a Data

play06:19

Center and consolidate things like

play06:21

cooling um bring their power costs down

play06:24

um and as as I've mentioned you know um

play06:28

you know they can do things that you can

play06:30

only do if you got the gpus together

play06:31

like you know give them access to each

play06:33

other's memory um so you know in the end

play06:36

like they're going to win on on on

play06:38

features and capabilities and and price

play06:42

so um we're very bullish um about you

play06:47

know blockchain and AI but we think the

play06:51

magic combination is when you run the AI

play06:53

on the blockchain because then you you

play06:55

start getting these um features and

play06:58

capabilities uh that you can't get on on

play07:01

centralized clouds so we think that's

play07:04

that's where blockchain adds value all

play07:06

right awesome so then let's dive into

play07:09

exactly what those features and special

play07:12

capabilities um on the blockchain site

play07:13

are obviously what is then so important

play07:17

about AI smart contract specifically

play07:21

well when you run AI as a smart contract

play07:24

you know it gains the properts of smart

play07:27

contract software so there are lots of

play07:29

really valuable properties actually but

play07:30

the three main ones are being tamper

play07:34

proof being Unstoppable and optionally

play07:37

being autonomous obviously those are

play07:39

pretty technical terms so let's break

play07:41

them down a bit can you explain tamper

play07:43

proof please well so tamper proof means

play07:46

that um some software is guaranteed to

play07:50

execute correctly right and you know

play07:53

when you run the software it always runs

play07:55

the correct logic against the correct

play07:58

data um and

play08:00

another way looking at it is that it

play08:02

means the software instances immune to

play08:05

Cyber attack and and that's going to be

play08:07

very important because um AI is going to

play08:10

become as big a target for hackers as

play08:14

crypto but tokens off tent so sort of an

play08:17

inherent security feature uh what about

play08:20

Unstoppable well unstoppable means that

play08:24

um it's guaranteed to run okay so you

play08:27

know by contrast if you're running AI on

play08:29

current it um the AI might crash um uh

play08:34

the the computer is running on might be

play08:35

rebooting the data center it's in might

play08:38

go down so that's a very different

play08:40

proposition um if you run AI as a

play08:42

smallart contract it's it's guaranteed

play08:44

to run and and that's going to be

play08:45

incredibly important because we're going

play08:47

to come to depend on on AI all right and

play08:52

let's finish things off with a third one

play08:54

autonomy so when a smallart contract is

play08:57

made autonomous it means that it cannot

play09:00

be you know modified or

play09:02

updated by anyone or um it can only be

play09:07

modified and updated by a dow and this

play09:11

is going to be very important for things

play09:13

like um smart wets because you know if

play09:16

you have a smart wallet with AI inside

play09:19

and it's not autonomous um somebody

play09:21

could modify it and steal your tokens

play09:24

all right I'd love to dive into a bit

play09:26

more of the actual practical application

play09:28

side of things so can you give me some

play09:30

examples maybe of how AI smart contracts

play09:33

might be used actually in

play09:36

practice well well let's start with um

play09:39

an example outside web 3 okay um I'll

play09:43

I'll start off with uh an imaginary law

play09:46

firm with a 100 lawyers right and um you

play09:51

know it's got a bunch of clients who are

play09:53

battling their way through through the

play09:56

courts and um let's say today you know

play10:00

when one of these lawyers has to prepare

play10:02

for

play10:02

trial um it it takes them three weeks

play10:06

right three weeks of work looking back

play10:08

at their previous legal filings case law

play10:11

um and so

play10:13

on um and let's imagine that this more

play10:16

firm in installs a kind of custom Oracle

play10:20

um Ai and and it's based on llama and

play10:23

it's keeping its information in an

play10:25

Associated Vector database CU that's

play10:28

what AI needs

play10:30

um this and and that this AI ingests you

play10:35

know um all of the previous

play10:38

Communications with their clients all of

play10:39

the legal filings they've made um all of

play10:43

you know the relevant Cas law um

play10:46

important legal test texts and and so on

play10:50

and now all of a sudden instead of

play10:52

taking like three weeks to prepare for

play10:54

trial the lawyer can do it in in 3 days

play10:58

and not only that um they can be very

play11:02

sure that they haven't missed something

play11:04

because the AI doesn't miss a thing so

play11:08

that's that's a huge Advance right

play11:10

because um you know if it takes two

play11:12

weeks to prepare for trial well um

play11:15

that's you know very expensive for their

play11:18

client and in addition to that you know

play11:21

when they're doing it themselves just

play11:22

reading through all the stuff they could

play11:24

miss something important so this is this

play11:27

this kind of thing will be absolutely

play11:29

brilliant but um let's imagine that this

play11:35

AI uh they're running is on current it

play11:40

okay the first problem we can see is

play11:43

that it's not tamperproof

play11:44

and it's going to be like a hot wallet

play11:47

for all their clients most important

play11:50

data that analogy yeah and um it's going

play11:54

to be targeted by hackers and and

play11:56

imagine if a hacker gets onto the

play11:58

infrastructure running that

play12:00

AI they're going to be able to

play12:02

exfiltrate all their clients most

play12:04

sensitive data and this will probably

play12:08

ruin their clients lives um cause them

play12:11

to lose

play12:12

their law lawsuits and probably ruin the

play12:16

law firm it'll go out of business um so

play12:21

very clearly it's important that this AI

play12:24

is Tamp proof and the hackers can't get

play12:26

onto it and exfiltrate all this

play12:27

sensitive data because it's otherwise

play12:30

it's going to be a hot wallet for

play12:31

sensitive

play12:32

data um in addition to that imagine you

play12:35

know um the lawyer you know Waits and

play12:38

some lawyer Waits until 3 days before

play12:40

trial because he's expecting yeah he's

play12:43

expecting that he can you know use this

play12:45

AI to prepare and so it's only going to

play12:47

take three three days otherwise it would

play12:49

take three weeks right so you know three

play12:51

days before um trial he he he sort of

play12:55

logs on and tries to access the AI and

play12:57

he finds that it's down maybe the AI has

play13:00

been encrypted by ransomware maybe the

play13:03

server machine that runs the AI um has

play13:06

gone pop uh maybe the data Center's down

play13:09

um he's going to be absolutely screwed

play13:12

um because he's come to depend on this

play13:14

Ai and all of a sudden it it's not

play13:17

running and and that's why this you know

play13:19

AI just has to be tamper proof and you

play13:22

know by running this AI as an internet

play13:24

computer smart contract um the law firm

play13:27

can be sure a that it doesn't get hacked

play13:30

and corrupted by a hacker or the data

play13:32

exfiltrated or encrypted by ransomware

play13:34

or the infrastructure goes down and it's

play13:36

not running when they need it um and

play13:38

this this is why and you know I could

play13:40

repeat that with same reasoning with

play13:42

lots of different examples this is why

play13:44

we believe um yes there are going to be

play13:47

millions or even billions of custom AIS

play13:49

running in the future but a large

play13:52

proportion of them are going to be

play13:53

running eventually as smart contracts

play13:56

because um people are going to need

play13:58

these guarantees that they can't be

play14:00

hacked and they're always going to be

play14:01

available all right I'd love to look at

play14:04

another example that you touched on

play14:05

earlier from specifically the web 3 side

play14:07

of things and that's of in the form of

play14:08

the smart wallet right

play14:10

yeah yeah so um web free like you know

play14:15

has the same kind of concerns that um

play14:18

the lawyers do first of all and they

play14:19

don't want their um AIS to be hacked and

play14:23

the data exfiltrated and they they want

play14:24

them always to be available espec

play14:26

especially if they're being used in

play14:27

conjunction with other smart tracks but

play14:30

they have this you know web 3 has this

play14:32

additional need for autonomy so um you

play14:36

know imagine a smart wallet so on the

play14:38

internet computer uh a smart contract

play14:41

can be a wallet right because um it it's

play14:45

a a multi-chain wallet um because it's

play14:47

able um to hold tokens on different

play14:49

chains using something called chain key

play14:51

and smart contracts on the internet

play14:53

computer can process web so they can you

play14:56

know create a web experience and the

play14:58

user can interact directly with them

play15:00

through a web browser once they've

play15:02

authenticated to it to them so um you

play15:05

can see this kind of thing at o.com ois

play15:09

y.com that's an example of a of a smart

play15:11

contract wallet now in the future um

play15:16

something like

play15:17

o uh might have an AI embedded right

play15:21

because it's just a small contract so

play15:23

therefore it can have an AI model in

play15:24

sight and this will mean that you can be

play15:27

on the beach right

play15:29

and uh pick up your phone and speak to

play15:33

your smart wallet and you'll be able to

play15:34

say something like cancel all previous

play15:38

instructions um if Bitcoin hits

play15:41

$100,000 uh sell half for usdc if eth

play15:46

falls 10 10% in a day by you know $500

play15:51

worth um and every night uh you know Buy

play15:55

$100 worth of ICP or something something

play15:57

like that and this is going to be very

play15:59

very cool because your wallet's going to

play16:00

have intelligence it's going to be like

play16:02

your personal

play16:04

broker um but to make this work you need

play16:07

autonomy because if that smart wallet

play16:11

wasn't autonomous that would mean that

play16:13

some developer controlled it or you

play16:17

know company controlled it so you know

play16:20

if a developer controls the smart wallet

play16:22

they can change the code and they can

play16:24

change the code so um it exports all

play16:26

your tokens um typically to their one or

play16:29

something like that right they can rug

play16:31

pull you or Rob you or whatever and um

play16:34

that's obviously very bad so the way to

play16:35

solve that is you is to C my wallets

play16:40

where you know the code can't be updated

play16:42

at all um or or or it's only know

play16:46

upgradeable by a dow or something like

play16:47

that and so the you know the wallets are

play16:49

autonomous nobody controls it there's no

play16:51

back doors um and you can repeat that

play16:53

example again you know across lots of

play16:55

things in web 3 and another example

play16:57

would be you you could imagine

play16:59

an internet computer aii smart contract

play17:02

that audits solidity you know solidity

play17:06

defi contracts on ethereum and if it

play17:08

sees that you know there are no bugs uh

play17:11

there are no back doors or other SEC

play17:13

security flaws you know kite marks them

play17:15

and says hey this um AUM smart contract

play17:18

is safe to use but again like you know

play17:21

unless it's autonomous um how could you

play17:24

trust that kite Mark for example if I

play17:27

personally ran that Ai and I could

play17:28

update it well I I could have it kite

play17:32

mark my own malicious solidity smart

play17:34

contracts on ethereum to steal

play17:35

everyone's ether or something so um you

play17:38

know autonomies the autonomy property is

play17:40

going to be very very important within

play17:41

web 3 all right so that Golden Triangle

play17:44

between tamper proove Unstoppable

play17:46

autonomous um of course you've done a

play17:49

number of demos fairly recently that

play17:51

demonstrated sort of the AI and smart

play17:53

contract capabilities on the internet

play17:56

computer specifically the most recent

play17:57

one focused specific spefically on

play17:59

facial recognition can you maybe tell me

play18:01

a bit more about that yeah so uh the

play18:04

reason we we chose to do a facial

play18:08

recognition demo um is it sort of brings

play18:12

home quite how powerful these AI smart

play18:15

contracts are um firstly like you can

play18:18

actually um have one doing facial

play18:22

recognition with practical performance

play18:24

right um

play18:27

and we we wanted to do that partly

play18:30

because we wanted to show how how how

play18:33

far we've

play18:34

Advanced um you know the performance of

play18:37

AI smart contract so when we did the now

play18:39

when we did the facial recognition demo

play18:41

um it was running probably like almost

play18:44

10 times faster than the first AI smart

play18:47

contract demo which had only been done

play18:49

um a few months before and um that

play18:53

reflected that just a huge amount of

play18:55

work that's been done behind the scenes

play18:57

a definity you know working away making

play19:00

improvements uh adding simmed thing

play19:03

called simmed instructions which allow

play19:05

multiple instructions to be executed in

play19:07

parallel to the execution environment

play19:09

and you know proposing those updates to

play19:11

the network nervous system the Dow that

play19:13

runs the internet computer and getting

play19:14

them into the

play19:15

network and uh you I wanted people to

play19:19

see uh the results of of all that work

play19:24

um but as it turns out um you know there

play19:27

are tons of app ation of uh smart you

play19:31

know facial recognition by smart

play19:33

contracts and um you know a lot of

play19:36

people are now playing with it and some

play19:38

you know have have reached out to us for

play19:41

support and guidance on on how how it

play19:44

canus and things they should do so uh

play19:47

you know that's been a huge success

play19:48

story ni some exciting things in store

play19:51

now of

play19:52

course I would be Miss if I didn't ask I

play19:54

think everyone wants to see large

play19:56

language models on the blockchain

play19:59

where are we on that um yeah I mean I've

play20:02

said many times you're going to be able

play20:04

to have a chat with with with the

play20:06

blockchain um it's coming in in fact

play20:09

there are very cut down large language

play20:11

models uh on the internet computer um

play20:15

already M um but you know we're focused

play20:18

on getting like really sophisticated

play20:20

large language models like llama 3 onto

play20:24

um the internet computer so the primary

play20:27

current limitation um is that uh the

play20:32

internet computer execution environment

play20:34

um runs small contracts that can have a

play20:37

maximum of 4 gigabyte

play20:39

of persistent main memory and that's

play20:43

because it's using 32-bit WM um you know

play20:46

a bit like Windows 95 was 32 bit back in

play20:48

the day I remember

play20:50

cloy so um uh soon uh we're going to put

play20:56

proposals into the net nervous system

play20:58

that will move the internet computer to

play21:02

64-bit wasm and that will make it

play21:05

possible to increase um the amount of

play21:09

persistent main memory that smart

play21:10

contracts have as needed um and was at a

play21:13

hoping we can um you know increase it

play21:17

something like 15 gig which will enable

play21:21

um you to fit like you know for example

play21:24

a 78 gab distillation of of laa 3 in

play21:31

into a smart contract which will be

play21:33

absolutely Mega um because that's a very

play21:37

sophisticated large language model um so

play21:41

obviously there's a lot of work behind

play21:42

the scenes uh you know defini is um you

play21:46

know a big player in in the wasm web

play21:49

assembly ecosystem actually the um one

play21:54

of the you know first few team members

play21:55

of guy Andreas rosberg was a co-inventor

play21:58

of um web assembly and you know that

play22:02

helps um it it's also the case that um

play22:06

you know increasing the amount of u

play22:09

persistant M memory that small contracts

play22:10

have requires a lot of other protocol um

play22:15

updates and improvements and so you know

play22:17

defini is working on that we'll put that

play22:19

into the network nervous system too um

play22:21

it's it's a non-trivial thing but but

play22:23

we're going to get there um and we're

play22:25

going to be able to run large language

play22:27

models on on the internet computer um

play22:31

and once we've done that I think the the

play22:34

the next um big Master for AI will be

play22:37

gyron which which involves um getting

play22:40

gpus onto no machines um which will take

play22:44

us a whole way further with respect to

play22:48

Performance and efficiency or allbe it

play22:50

you know the reason that you run AI as a

play22:53

SM contract isn't isn't because it's the

play22:55

most efficient way of running an AI

play22:57

right the reason you do that is because

play22:59

you want it to be T proof you want it to

play23:00

be unstoppable uh and in the case of a

play23:02

wallet or something like that you want

play23:03

it to be

play23:04

autonomous all right and to finish us up

play23:07

there's been a lot of chatter on X

play23:10

largely initiated by you regarding

play23:13

something other than AI smart contracts

play23:15

sort of another Advance if you want to

play23:17

call it that that the internet uh

play23:18

computer has in the pipeline it's been

play23:21

teased I'm going to ask cheely can you

play23:25

speak in your cryptic ways about that a

play23:29

bit well you know I can't tell you what

play23:32

it is I mean um we're aiming to get this

play23:34

out in in early September um in beta

play23:37

form but people will be able to use it

play23:38

and you know I'm itching to talk about

play23:40

it I'm in itching to show it to people

play23:42

I'm so excited and it's it's got me you

play23:45

know probably more excited than uh

play23:48

anything else for a very long time um

play23:51

and you know running uh AI on on

play23:55

blockchain is like the first big advance

play23:58

um the internet computer delivered uh in

play24:01

this space with respect to AI but

play24:03

there's another big Advance coming and

play24:05

when you see it um you realize that the

play24:09

future of tech is going to be completely

play24:11

different um and it's a real game

play24:15

changer I and you know what's really

play24:17

cool about this thing is it crystallizes

play24:19

so much of the work that we've done over

play24:21

years and years um creating this like

play24:24

novel uh third generation blockchain

play24:27

which hosts this secure Unstoppable

play24:29

autonomous um serverless code where you

play24:33

can literally build everything end to

play24:36

end um on on this decentralized

play24:39

network um and it's something again that

play24:44

the internet computer can do uniquely So

play24:47

currently as you know uh only the

play24:49

internet computer blockchain can run AI

play24:51

there's lots of blockchains do like

play24:53

tokens for AI and all this kind of stuff

play24:55

and there's only one blockchain that can

play24:57

run AI has a smallart contract on chain

play24:59

and that's the internet computer um and

play25:03

this is going to be a bit like that you

play25:05

know it's it's it's something that only

play25:07

the internet computer can do um and I

play25:11

think it shows you know closes the loop

play25:14

and and shows you how tightly woven AI

play25:18

is going to be with with the future of

play25:19

tech and it's one of those things that

play25:21

you look at and you just go wow um this

play25:24

this is incredible this is a game

play25:25

changer but people will have to wait I

play25:28

mean I I'm going to have to wait too

play25:29

because I'm desperate to talk about it

play25:31

and show people and get discussions

play25:33

going but but it's not not not a lot

play25:35

long now right like less than five weeks

play25:39

four and a half weeks I mean consider

play25:41

yourselves teased H I imagine people are

play25:43

spping their calendars right now I mean

play25:46

actually I I I I I canceled half my

play25:49

holiday but there me it be on holiday

play25:50

and the reason I did that is to make

play25:52

sure that this thing stays all schedule

play25:54

and get this out and we get it out so it

play25:58

it's that important um I'm super excited

play26:01

you know I'm just desperate to share

play26:02

with people but it'll have to wait it's

play26:04

not long now what a fantastic note to

play26:06

end on I think we can wrap it up with

play26:08

that D thanks for taking the time thank

play26:11

you all for uh thank you Tracy checking

play26:12

us out and looking forward to hopefully

play26:15

having exciting news to share within the

play26:17

next couple weeks already yep cool bye

play26:20

everyone bye everyone

Rate This

5.0 / 5 (0 votes)

Related Tags
ブロックチェーンAIインタビューセキュリティスマートコントラクトデцентр化インキュベーションウェブ3技術革新未来予測
Do you need a summary in English?